Announcements
New models, upcoming launches, paused services, price changes and retirements are posted here first. Price changes are announced at least one day ahead.
October 2026
-
Notice
NezhaGate is now KovAPI — new address kovapi.com
NezhaGate has been renamed KovAPI. The website now lives at kovapi.com, and the docs at docs.kovapi.com. · Your account, balance, API keys and prices stay the same. · The existing API address https://nezhagate.com/v1 keeps working permanently, so your code needs no changes. For new integrations you can use https://kovapi.com/v1. · Please sign in once on kovapi.com; old web addresses redirect automatically. · Images and videos you generated before keep their links. Questions: support@kovapi.com or our Telegram group. (Posted October 31, 2026 at 14:03 UTC / 22:03 Beijing time)
-
Notice
Compensation for recent video generation failures
Between October 17 and October 30, some video models (mainly Seedance 2.5) failed more often than usual. Failed jobs were never charged, but many of you had to retry again and again and wait for nothing, and we are sorry for that. Based on how much each account was affected during that period, we have given the most affected users 100–400 credits as compensation. The credits have already been added to your balance. We have also optimized our video generation service during this time, so generations should now be more stable. Thank you for your patience and understanding. Published 2026-10-31 13:33 UTC (2026-10-31 21:33 Beijing time)
-
New
Seedance 2.0 Mini is back and 2.0 Fast adds 480P, at a launch price
Seedance 2.0 Mini (seedance-2.0-mini) is available again, and Seedance 2.0 Fast (seedance-2.0-fast) now renders 480P too: · Both take 480P / 720P, 5-15 seconds and five aspect ratios, with reference images, videos and audio in the same request; every clip has sound. · Launch price: Mini 480P $0.022/s and 720P $0.046/s; Fast 480P $0.046/s; Fast 720P is unchanged at $0.075/s. The launch price runs until 2026-11-14 16:00 UTC (Nov 15 00:00 Beijing time); after that Mini is $0.024/s at 480P and $0.052/s at 720P, and Fast is $0.052/s at 480P. · With a reference video, its seconds count toward the billed length: a 5-second clip with a 7-second reference video is billed as 12 seconds. Reference images and audio are free. Failed generations are not charged and can be resubmitted as they are. Published 2026-10-31 09:43 UTC (2026-10-31 17:43 Beijing time)
-
Notice
A note on recent video quality: please tell us if something looks off
Over the past few days we have made a lot of changes to our video generation to keep Seedance and other video models available. With this many changes we may have missed things, and we apologize to anyone affected. Please contact us if a video you generated: · is clearly lower in resolution or sharpness than the tier you chose (for example, you picked 720P but got a blurry or smaller clip); · does not look like Seedance 2.5 output — style, motion or texture clearly off; · used a real-person reference but shows someone who does not match it, or a face that looks blurred, swapped or pre-processed. To protect your privacy, our team does not look at your prompts or videos without your permission, so we cannot catch these cases on our own — we rely on you. Send us the task ID (it starts with img_) and a short description; with your consent we will check that task's prompt and video, and if the problem is confirmed we will refund that task in full. Videos are kept for 7 days, so please report within that window. To reach us: send /support in our Telegram group, or email support@nezhagate.com. Thank you for your patience and for keeping us honest. We will keep working to make quality steady. Published: 2026-10-30 15:45 UTC (2026-10-30 23:45 Beijing time)
-
Notice
Seedance 2.5 per-second is back; the special tier is still under maintenance
Seedance 2.5 per-second (seedance-2.5) is back: 480P, 720P and 1080P jobs can be submitted again, at the same prices. The special tier (30-second clips) is still under maintenance; we will announce when it returns. The Token version remains available. Failed generations are not charged and can be resubmitted as is. Thank you for your patience. Published: 2026-10-30 15:23 UTC (2026-10-30 23:23 Beijing time)
-
New
Gemini 4 Argon Beta is back, now with Google Search
Gemini 4 Argon Beta is callable again (model ID: gemini-4-argon): · Three thinking levels: set reasoning_effort to low / medium / high; low is the default. · New web search: add "web_search": true and the model looks things up with Google Search, citing its sources (message.annotations). Each search costs $0.0035. Prices are unchanged: input $0.50, output (including thinking) $3.00, cache hits $0.05, all per million tokens. (Posted October 30, 2026 at 06:58 UTC / 14:58 Beijing time)
-
New
GPT Image 2.5 2K / 4K: transparent backgrounds, multiple references, 4:3 and 3:2
GPT Image 2.5 Flare and Sunburst now support, at 2K / 4K: · Transparent backgrounds (4K): send background: "transparent" to get a PNG with an alpha channel, for both text-to-image and image editing. · Multiple references: image editing takes up to 5 reference images, PNG or JPEG (4K used to accept PNG only). · 4:3 and 3:2: seven aspect ratios are now available — 1:1, 16:9, 9:16, 2:3, 3:4, 4:3 and 3:2. Prices are unchanged: 6 credits per 2K image, 10 credits per 4K image. Please note: · Output is PNG only. output_format: "jpeg" or "webp" is rejected at submission; if you need JPEG / WebP, convert the PNG on your side. · Each request returns one image (n=1); submit several requests for more images. · Mask inpainting is not supported yet, and 2K does not support transparent backgrounds yet. Unsupported parameters are rejected at submission and never charged. (Posted October 30, 2026 at 05:46 UTC / 13:46 Beijing time)
-
Notice
Top-up bonus changes now in effect
From October 30, 2026 at 02:02 UTC (10:02 Beijing time), new top-ups are credited as follows: · $50 (CNY 360): 10,300 credits with WeChat Pay, card or USDT · $500: 107,000 credits by card or USDT · $1,250: 275,000 credits by card, 280,000 credits with USDT The $5, $15 and CNY 10 packs are unchanged; WeChat Pay is still unavailable for the two largest packs. Orders created earlier are credited as shown when they were placed. The USDT $500 pack gave 110,000 credits from 02:02 to 02:24 UTC (10:02 to 10:24 Beijing time); from 02:24 UTC (10:24 Beijing time) it gives 107,000 credits, the same as by card. The refund fee policy is unchanged: top-ups made from October 28, 2026 at 16:00 UTC (October 29, 00:00 Beijing time) carry a $1.40 fee per refund, earlier top-ups $0.40. Unused purchased credits can be refunded within 7 days of a top-up; bonus credits are not refundable. (Posted October 30, 2026 at 02:03 UTC / 10:03 Beijing time; updated 02:24 UTC / 10:24 Beijing time)
-
Notice
Kovapi domain and rebranding preview: planned for 1 November
We plan to launch the new domain kovapi.com and adopt the Kovapi brand name throughout the service on 1 November 2026. The website, payment entry points, payment pages and other displays and descriptions containing the brand name will be updated to Kovapi. Please be aware that the name shown on these pages may change at that time. The rebranding will be introduced gradually in stages, with uninterrupted normal use as a requirement. You can continue using the service as usual and do not need to change your configuration because of this advance notice. If any settings changes are needed later, we will notify you ahead of time. This is advance notice of the domain and name changes, not a confirmation that the transition is already complete. Please refer to the formal notice for the exact launch time and transition arrangements. Any changes to the plan will be announced separately. Thank you for your support and understanding. Published: 2026-10-29 15:04:13 UTC
-
Notice
Service notice: NSFW sexual content is prohibited
Do not use this website or its API to generate, rewrite or edit NSFW sexual content. This notice applies to all models and to text, image, video and audio generation. Prompts, uploaded materials and generated outputs must all comply. Prohibited content includes explicit sexual acts, nudity or intimate body-part displays intended for sexual arousal, erotic roleplay, and pornographic audio or video; any sexualised content involving minors or people who appear to be minors; sexual violence, sexual exploitation, non-consensual intimate imagery, and pornographic composites using a real person’s likeness. Fictional, anime, face-swap or “personal use only” framing does not exempt such content. Do not bypass review through coded language, phonetic substitutions, split prompts, image editing or switching models. If content is rejected, stop submitting prohibited material. An occasional successful generation does not mean the content complies with our rules. Ordinary non-pornographic depictions of people, clothing, relationships, art, education or medicine are not automatically prohibited because of their subject matter. These labels must not be used to disguise pornography, and each model’s content requirements still apply. Ensure you have the right to use uploaded materials and respect others’ privacy, likeness rights and other lawful rights. Help maintain a responsible creative environment. Published: 2026-10-29 14:01:17 UTC
-
Notice
Seedance 2.5 per-second and special-price versions: under maintenance; please do not attempt generation
The per-second and special-price versions of Seedance 2.5 are currently under maintenance. Please do not attempt generation. Review of submitted materials has recently become stricter, and we are working to address the related issues. Maintenance is expected to be completed on 30 October 2026, subject to actual recovery. We apologise for the inconvenience and thank you for your patience and understanding. Published: 2026-10-29 12:19:25 UTC
-
Notice
Card top-ups restored
From October 28, 2026 at 16:00 UTC to October 29 at 02:36 UTC, card payments were temporarily unavailable for some top-up packages. Card top-ups are now available again. Please refresh the top-up page and try again. We sincerely apologize for the inconvenience and thank you for your patience and understanding. Published: 2026-10-29 02:44 UTC
-
Notice
Seedance review is stricter recently: avoid repeated identical submissions
Seedance content review has recently become stricter, increasing the likelihood of generation failures due to content review. If a request fails review, please avoid repeatedly submitting the same prompt and reference materials. Review and adjust your prompt or references first, or try another available version. Review results may differ across versions, but switching versions does not guarantee approval. Please follow the content rules. Published: 2026-10-28 14:49:44 UTC
-
Notice
Kling maintenance: temporarily unavailable
Kling is currently under maintenance and temporarily unavailable. It has been temporarily removed from the available models. We will announce when service resumes. Published: 2026-10-28 14:19:15 UTC
-
Notice
Wan 3.0: Standard 480P / 720P / 1080P available; Prime paused
Wan 3.0 Standard at 480P, 720P and 1080P is available. Wan 3.0 Prime is not available for now; we will announce when it returns. Until then, Prime requests are refused at submission and are not charged. Updated: 2026-10-26 15:21 UTC (23:21 Beijing time). First published: 2026-10-24 10:23 UTC.
-
New
Kling 4.0 is live: reference images, videos and audio, 3-30 seconds, billed per second at 720P / 1080P
Kling 4.0, the next-generation video model from Kuaishou's Kling, is now available in the Playground and through the API. · No references for text-to-video, or up to 10 reference images (characters and settings), 5 reference videos (motion and camera) and 3 reference audio clips (a character's voice) together · Any length from 3 to 30 seconds, 720P / 1080P, landscape 16:9 or portrait 9:16 · Every clip has sound: write dialogue, effects and ambience into the prompt Billed per second: 720P $0.09/s (18 credits), 1080P $0.125/s (25 credits). With reference videos the billed seconds are the output seconds plus the reference-video seconds: 10 s of 720P output with a 5 s reference video bills 15 s, 270 credits. Reference images and audio are free; failed renders are refunded in full. Requirements: each reference video 2-30 s and at least 700 px high (e.g. 1280x720); each reference audio 5-30 s of clear human speech. A file outside these limits is refused at submit, not charged. Model ID: kling-4.0. Parameters and examples are on the model page and in the API docs. Enjoy! (Published October 26 03:53 UTC / 11:53 Beijing time; updated 05:09 UTC / 13:09 Beijing time: reference video, reference audio and up to 30 seconds are open)
Kling 4.0 -
New
GPT Image 2.5: five aspect ratios for 2K and approximately 8MP 4K generation and editing
Following several test rounds, GPT Image 2.5 Flare and Sunburst now offer five aspect ratios for both text-to-image generation and single-reference editing at 2K and approximately 8MP 4K: 1:1, 16:9, 9:16, 2:3 and 3:4. 4K delivers approximately 8 million pixels. Actual dimensions vary with aspect ratio and may differ from the requested width and height. Each request produces one PNG. 2K edits accept one PNG or JPEG reference; 4K edits accept one PNG. Multiple outputs, multiple references, strict mask editing, transparent backgrounds and JPEG/WebP output remain unavailable. Across 57 additional requests, we reviewed 42 originals (including one reused sample), checking dimensions and visual results. Failures and unknown outcomes were not counted as successes. Most combinations have delivered samples; some newly opened combinations still need observation. Requests can be slow or fail, and no fixed completion time or stability guarantee is offered. Give it a try and let us know what you find! Published: 2026-10-24 16:43:09 UTC
-
New
Seedance 2.5 Token is live: reliable generation, fast results, and draft mode
Seedance 2.5 Token is now available for reliable video generation and fast results! Draft mode is also live: preview a 480p draft, then use that draft task to generate a 1080p final video. Image, video, and audio references are supported through the Playground and API. Pay for actual token usage, at approximately 95% of BytePlus international public API reference pricing. The comparison varies slightly by billing tier; see the model page for exact rates. Drafts and final videos are billed separately. Give it a try!
-
Notice
Seedance 2.5: 1080p available again
1080p requests for Seedance 2.5 (seedance-2.5, billed per second) are available again through the API and Playground. For an existing task, check its original job ID before resubmitting. Seedance 2.5 · 30s remains under maintenance and is not included in this recovery. Thank you for your patience. 2026-10-24 08:34 UTC
-
Notice
Seedance 2.5 service restored
Seedance 2.5 (seedance-2.5) is available again through the API and Playground. For an existing task, check its original job ID before submitting again. Seedance 2.5 · 30s remains under maintenance and is not included in this recovery. Thank you for your patience. 2026-10-24 06:22 UTC
-
Paused
Seedance 2.5 temporarily unavailable due to a widespread outage
Seedance 2.5 is currently experiencing a widespread outage and is temporarily unavailable. Please hold off on submitting new tasks and avoid repeated retries. There is no confirmed recovery time yet. We will post a separate announcement when service is restored. We apologize for the inconvenience. Published: 2026-10-24 00:21:06 UTC
-
Notice
Seedance 2.5 1080p maintenance — submissions remain open
Seedance 2.5 at 1080p is currently under maintenance and generation failures are more frequent. Submissions remain open, but we recommend waiting for service stability for non-urgent tasks. Failed generations are not charged; reserved credits are returned. If your client times out or does not show a result, check the original task before submitting again. A recovery update will follow. Published: 2026-10-23 03:14:25 UTC
-
Notice
Flare / Sunburst 2K image usage update
GPT Image 2.5 Flare and Sunburst now support the following 2K workflows: • Text-to-image: 16:9 landscape or 2:3 portrait, one output per request. • Editing: one PNG reference and one output; multiple references and masks are not currently supported. • In Playground, select 2K and one of these aspect ratios. API examples: 2048×1152 landscape or 1360×2048 portrait. Quality is fixed at medium. • Original images meeting the 2K requirement are returned. Actual dimensions may be larger, such as 1536×2304 portrait. Use the returned width and height; images are not enlarged or cropped merely to match a requested size. Other aspect ratios, transparent backgrounds and non-PNG output are outside this update. Busy requests may wait in the queue; check the original task instead of submitting duplicates. This update does not change 1K or 4K. For API calls set resolution="2K" and n=1. Editing uses Content-Type: application/json with images:[{"image_url":"PUBLIC_PNG_URL"}]; direct multipart uploads to this gateway editing endpoint are not currently supported. Published: 2026-10-22 16:50:55 UTC.
-
Notice
Sonnet 5.5 service incident: repair in progress
The Sonnet 5.5 chat service is experiencing an incident. We are investigating and working on a fix. We will announce the recovery time separately. We apologize for the inconvenience and appreciate your patience. Published: 2026-10-22 11:29:03 UTC.
-
Notice
Seedance service restored
The Seedance video generation incident has been resolved and service has been restored. Thank you for your patience and understanding. Published: 2026-10-22 11:02:38 UTC.
-
Notice
[Resolved] Seedance service incident history
This incident has been resolved. See the recovery announcement for the current status. The following is the historical notice published during the incident: The Seedance video generation service is experiencing an incident. We are investigating and working on a fix. We will announce the recovery time separately. We apologize for the inconvenience and appreciate your patience. Published: 2026-10-22 10:29:07 UTC.
-
Notice
Generated results will be kept for 7 days: please download in advance
From 2026-10-29 09:49:40 UTC, the retention period for generated images and videos will change from 15 days to 7 days. The new policy has a full 7-day notice period. Until the effective time, normal cleanup under the current 15-day policy continues. After it takes effect, files generated more than 7 days earlier will be removed, including existing files. The 7-day policy will then apply to ongoing cleanup, and expired image and video links will stop working. Uploaded reference files will still be kept for 7 days. Existing retention periods for call logs, text content and database backups remain unchanged. Please download any images and videos you want to keep before the effective time. Thank you for your understanding and support. Published: 2026-10-22 09:49:40 UTC.
-
Retired
GPT 5.5 retired and Gemini service changes
GPT 5.5 was retired on 2026-10-22 at 14:00 UTC. It has been removed from the available-model list and no longer accepts new requests. The Gemini 4 family (Gemini 4 Argon, including Low, Medium and High) is temporarily unavailable. Reopening plans will be announced separately. Gemini 3.1 Pro is temporarily suspended with immediate effect. Its model page and documentation are retained. Whether and when it will reopen will be announced separately. The Gemini 3.6 Flash family (standard, High, Low and Tiered) is temporarily suspended with immediate effect. Reopening plans will be announced separately. Gemini 3.5 Flash was already discontinued. For Gemini chat, you can switch to Gemini 3.8 Flash. Thank you for your understanding and support. Published: 2026-10-22 04:26:18 UTC.
GPT-5.5Gemini 3.1 ProGemini 3.6 FlashGemini 3.6 Flash HighGemini 3.6 Flash LowGemini 3.6 Flash TieredGemini 3.5 FlashGemini 4 Argon -
Price change
Advance notice: Sunburst 2K pricing adjustment
Thank you for your continued support. Due to increased costs, and to help maintain and improve service reliability, Sunburst 2K text-to-image generation and image editing will change from 4 to 5 credits per image on 2026-10-20 13:11:15 UTC. Until then, Sunburst 2K remains 4 credits per image. Prices for 1K and 4K will remain unchanged. We are sorry for any inconvenience this adjustment may cause. We will keep working to improve the service, and we appreciate your understanding and support. Published: 2026-10-18 13:11:15 UTC.
-
Paused
Sunburst / Flare 2K: service disruption
2K generation and editing requests for GPT Image 2.5 Sunburst and GPT Image 2.5 Flare are currently experiencing service failures. Please pause submissions and retries until we announce recovery. Do not resubmit tasks that are still processing. Failed tasks are not charged, and any credits reserved for failed tasks are automatically returned. We are investigating and will announce recovery separately. A recovery time has not yet been confirmed. We apologize for the inconvenience and thank you for your patience. Published: 2026-10-18 07:28:28 UTC.
-
Price cut
DeepSeek V4.1 Flash: half the official off-peak price, all day
DeepSeek V4.1 Flash now has one lower rate around the clock, with no peak-hour surcharge. All prices are USD per 1M tokens: Input (cache miss): $0.075 Output (including reasoning tokens): $0.30 Cache reads: $0.0015 All three are 50% below DeepSeek's current official off-peak rates of $0.15 / $0.60 / $0.003, and 75% below its peak rates. Only actual cache hits receive the cache rate; misses use the input rate, with no separate cache-creation premium. Keep using deepseek-v4.1-flash. Try your everyday tasks over the next two days and let us know how response speed and answer quality feel. Effective and published: 2026-10-17 16:30:05 UTC.
-
Notice
DeepSeek V4.1 Flash updated to the 1017 release
DeepSeek V4.1 Flash has been updated to the 1017 release, published on October 17. This release focuses on coding and agent tasks, with improved task efficiency. Keep using deepseek-v4.1-flash. The API format and prices remain unchanged. Thinking controls, tool calls and long context are supported; repeated prefixes are billed according to the cache tokens actually hit. Try the updated DeepSeek V4.1 Flash. Published: 2026-10-17 15:01:01 UTC.
-
Notice
Grok image editing: up to 5 references on Standard, 3 on Quality
Input reference-image limits differ by tier: Standard, grok-imagine-image, currently accepts up to 5; Quality, grok-imagine-image-quality, currently accepts up to 3. The generic “up to 8 reference images” wording in our existing documentation does not apply to either tier. Please follow this notice. We completed a real 1K Standard edit with 5 references, and all five reference shapes appeared in the result. Sending 7 references returns HTTP 400. Generated text and fine details may still contain errors, so check important content. Current input-count errors for the Quality tier explicitly state a limit of 3. An error such as “This model supports at most … input image(s), but … were provided.” means that the input references exceed the selected tier’s limit. It is unrelated to the output count n. Reduce the number of references before submitting again; do not repeatedly resend the unchanged request. Use image for one reference or images for multiple references. Choose one format and avoid submitting the same image in several fields. Published: 2026-10-17 10:20:44 UTC.
-
Notice
Seedance 2.5: reference-audio duration guidance
For Seedance 2.5 reference audio, we recommend about 2–10 seconds per clip and no more than 30 seconds in total across all reference audio clips. A single clip longer than 10 seconds can increase the chance of processing failure. This is reliability guidance, not a new 10-second hard limit. Longer audio within the existing parameter limits can still be submitted; this guidance does not block, trim or automatically split it. Shorter reference audio does not guarantee a successful generation. If a job returns a generic content_policy / invalid_request_error without identifying the review item or field, reference-audio duration is one item to investigate. That response alone does not establish that a particular asset violates policy, or that every failure is caused by long audio. When a specific review reason is provided, follow that explanation. Whether multiple audio clips can preserve a complete song continuously still needs confirmation. This guidance is not a verified automatic-stitching workflow. Keep your original audio intact. If failures continue, contact support with the job ID and avoid repeatedly submitting the same request. We are sorry for the inconvenience and will continue improving processing reliability and error messages. Published: 2026-10-17 05:33:20 UTC.
-
Notice
[Historical] Claude 5.5 service recovery on October 17
This is the historical recovery notice published on October 17. See the latest incident notice for the current Sonnet 5.5 status. On October 16, 2026 (UTC), an unexpected service disruption affected some requests to Claude Opus 5.5 and Sonnet 5.5. Both models are now available again. You can resume making requests. We apologize for the inconvenience and thank you for your patience and understanding. Published: 2026-10-17 03:54:54 UTC.
-
Notice
Seedance 2.5 · 30s temporarily closed; reopening after an upgrade
To improve generation reliability and the overall experience, Seedance 2.5 · 30s (the special-price 30-second option) is temporarily closed to new web and API submissions. We will reopen after an upgrade and verification, with the timing announced separately. Accepted tasks that are still processing will continue through the existing process. Please do not submit duplicates. This pause applies only to seedance-2.5-30s. We apologize for the inconvenience. Thank you for your patience and support. Published: 2026-10-16 16:33:38 UTC.
-
Notice
Seedance 2.5: issues during the service update
On October 15, 2026 (UTC), some Seedance 2.5 video-generation requests experienced 504 timeouts, connection issues or longer waits. Some 504 errors may have been related to connection issues during the service update. We are continuing to investigate the causes of individual failures. We are working to improve service stability and error messages. Jobs confirmed as failed are not charged, and reserved credits are returned automatically. If you encounter an issue, keep the job ID or Call Logs record and contact support. Please avoid submitting the same request repeatedly in quick succession. We apologize for the inconvenience to affected users and for not providing a clear explanation sooner. Our team will continue following up and share progress through announcements. Thank you for your understanding and patience. Published: 2026-10-16 06:49:27 UTC.
-
Notice
Seedance 2.5 reference images and failure messages
If a Seedance 2.5 job fails, check the failure details in Call Logs, the video-generation page, or the error field in the job-status response. For asynchronous jobs, also check status: an HTTP 200 polling response can still contain status=failed. If a generic content_policy or invalid_material error does not identify the review item or material field, an oversized reference image may be one factor affecting processing. For material that complies with the content rules, keep the original and prepare a compressed copy, reducing pixel dimensions as appropriate while preserving the character's appearance. Aim for around 2 MB or less per image as troubleshooting guidance. The specific cause still needs to be checked against the job records. Treat 502, 503, 504, connection drops and timeouts according to the service-error message; these do not establish that an image is too large. If failures continue, keep the job ID and contact support. We are sorry for the inconvenience. Our team will continue improving processing reliability and error messages. Thank you for your understanding and patience. Published: 2026-10-16 04:14:04 UTC.
-
Notice
GPT Image 2 series: all 1K image generation services restored
1K image generation has been restored for GPT Image 2, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst. All three are available again. We apologize for the inconvenience caused by the interruption. Thank you for your patience and support. Published: 2026-10-16 00:08:23 UTC.
-
New
Gemini 4 Argon Beta is available
Gemini 4 Argon is available in Beta for text chat and streaming. Try it on the model detail page. Model ID: gemini-4-argon. Choose Low, Medium or High with reasoning_effort; Low is the default. All levels use the same rates, and thinking tokens are billed as output. Per million tokens: input $0.50, output $3.00, cached input hits $0.05. The official 1,048,576-token window is shared by input and output. We verified retrieval with 349,709 input tokens. Repeated long inputs produced implicit cache hits, billed at $0.05 per million tokens. Hits on every repeat are not guaranteed. Million-token output is outside this Beta offering. A hard max_tokens output cap was not demonstrated; limit answer length and split long jobs. Details and playground: /en/model/gemini-4-argon Published: 2026-10-14 07:05:01 UTC.
-
Notice
Gemini 3.1 Pro is temporarily unavailable
Gemini 3.1 Pro is currently unavailable, and requests may fail. We are investigating. There is no confirmed recovery time; we will post an announcement when service is restored. The confirmed failed requests in this incident were not charged. We apologize for the inconvenience and thank you for your patience. Published: 2026-10-13 10:51:54 UTC.
-
Notice
[Historical notice] Sunburst 2K and 4K: unstable service, maintenance in progress
The following records an earlier service state. Sunburst / Flare 2K is currently experiencing disruption; please pause retries. Latest 2K notice: /en/announcements#2026-10-18-sunburst-flare-2k-service-disruption The 2K and 4K options of GPT Image 2.5 Sunburst are currently unstable, with an approximate success rate of 60%. Maintenance is in progress; we will post an update when service stability is restored. Requests may still fail during maintenance. Please do not duplicate tasks that are already processing. We apologize for the inconvenience and thank you for your patience and support. Published: 2026-10-13 04:27:14 UTC.
-
Notice
30-second special option: long prompts, wait times and content review
You can now submit prompts of around 10,000 characters to the Seedance 2.5 special-price 30-second option. Generation may still fail; success is not guaranteed for every request. Allow approximately 30 minutes for generation; actual wait times may vary. Please do not duplicate tasks that are still processing. Requests that violate content safety rules are automatically rejected during review. Please ensure your content meets these requirements. Thank you for your patience and support. Published: 2026-10-13 02:57:14 UTC.
-
Notice
Seedance 2.5 special-price 30-second option restored
The Seedance 2.5 special-price 30-second option is available again. You can submit new tasks through the website and API using the parameters shown on the model page. Please do not duplicate tasks that are still processing. Thank you for your patience and support. We apologize for the inconvenience. Published: 2026-10-13 02:16:06 UTC.
-
Notice
[Previous notice] Brief Seedance video service incident resolved
This is a previous status. The special-price 30-second option is available again. Latest notice: /announcements#2026-10-13-seedance25-30s-restored Seedance video generation experienced a brief disruption, which has now been resolved. We apologize for the inconvenience. The Seedance 2.5 special-price 30-second option remains paused as previously announced. Reopening will be announced separately. Published: 2026-10-12 23:44:01 UTC.
-
Notice
[Previous notice] Seedance 2.5 special-price 30-second option temporarily paused
This is a previous status. The special-price 30-second option is available again. Latest notice: /announcements#2026-10-13-seedance25-30s-restored The Seedance 2.5 special-price 30-second option is temporarily closed to new submissions. There is no confirmed reopening time; we will announce it when available. Tasks already submitted and still processing will continue as usual. Please do not submit duplicates. We apologize for the inconvenience. Check the website for updates. Published: 2026-10-12 14:08:16 UTC.
-
Notice
[Previous notice] Sunburst incident resolved
This notice records an earlier incident. Sunburst 2K and 4K are currently unstable and under maintenance. Latest notice: /announcements#2026-10-13-sunburst-2k4k-maintenance Some GPT Image 2.5 Sunburst image generation and editing requests failed during the recent overnight incident because of an upstream service fault. Some failures were incorrectly shown as invalid request parameters. We apologize for the disruption and confusion. The incident has been addressed and the related error-classification issue has been fixed. Affected failed requests were not charged, and reserved credits were refunded. You may resubmit failed tasks; please do not duplicate tasks that are still processing. Thank you for your understanding and support. Published: 2026-10-12 03:03:14 UTC.
-
Notice
[Previous notice] Special-price 30s videos: only 16:9 and 9:16, with prompt-length limits
This is a previous status. The special-price 30-second option is available again. Latest notice: /announcements#2026-10-13-seedance25-30s-restored This is a previous status. The special-price 30-second option is now paused. Latest notice: /announcements#2026-10-12-seedance25-30s-paused The Seedance 2.5 special-price 30-second option is currently available only in 16:9 landscape and 9:16 portrait. Other versions are not yet available; we will announce them when they open. Please keep prompts within 4,000 characters. Prompts exceeding 8,000 characters are rejected directly by the generation service. If you receive a length-limit error, shorten the prompt before retrying instead of resubmitting it unchanged. We are working to improve the service experience. Thank you for your understanding and support! Failed requests are not charged. Please avoid duplicate submissions while a task is processing. Availability and requirements for this option may change. Check the latest website announcements before submitting, or join our Telegram group for updates: https://t.me/+-erBoH9-AYY4MTM1 Published: 2026-10-11 07:02:55 UTC.
-
Notice
Sunburst 2K is now 4 credits per image
The 2K tier of GPT Image 2.5 Sunburst has been reduced from 5 to 4 credits per image, effective immediately for both 2K image generation and editing. 1K and 4K prices remain unchanged. We apologize for any inconvenience caused by the recent service adjustments. A targeted goodwill grant of 300 credits has been added directly to the relevant account(s); this is not an account-wide promotion. Thank you for your patience and support. We hope you enjoy creating with us. Please follow our announcements for the latest updates. Published: 2026-10-11 04:19:58 UTC
-
Notice
Claude Opus 5.5 and Sonnet 5.5 pricing update on October 12
Prices for Claude Opus 5.5 and Sonnet 5.5 will change on 2026-10-12 at 04:00 UTC. Current prices remain in effect until then. All amounts are USD per million tokens, in the order input / output / cache read / cache write: Opus 5.5: current $2 / $10 / $0.10 / $2.50; new $2.50 / $12.50 / $0.125 / $3.125. Sonnet 5.5: current $1 / $5 / $0.10 / $1.25; new $1.50 / $7.50 / $0.15 / $1.875. Thank you for your understanding and support. Please check our announcements for updates. Published: 2026-10-11 04:18:45 UTC
-
Notice
[Previous notice] Seedance 2.5 special-price 30s tier: portrait and landscape submissions open
This is a previous status. The special-price 30-second option is available again. Latest notice: /announcements#2026-10-13-seedance25-30s-restored This is a previous status. The special-price 30-second option is now paused. Latest notice: /announcements#2026-10-12-seedance25-30s-paused You can now submit Seedance 2.5 special-price 30-second tasks in 9:16 portrait and 16:9 landscape. For now, keep prompts to 4,000 characters or fewer. Prompts exceeding 4,000 characters will cause generation to fail. Generation times may vary. Please wait for the final task status and avoid duplicate submissions while a task is processing. We apologize for the inconvenience and thank you for your patience. Published: 2026-10-11 01:16:51 UTC.
-
Notice
Seedance 2.0 is available again
Seedance 2.0 is open for submissions again on the website and through the API. The 480P, 720P and 1080P resolution options are all enabled. Duration, aspect ratio and reference inputs remain available within their existing supported ranges. Generation times may vary. Check the final task status and avoid resubmitting while a task is processing. Thank you for your patience. Published: 2026-10-10 15:27:59 UTC.
-
Notice
Service statement: failed requests are free, videos refunded after 120 minutes
What we guarantee 1. Failures are free. A failed request costs no credits. Credits reserved when you submit an image or video are returned automatically if it fails. 2. Long video waits are refunded. If a video is not finished 120 minutes after you submit it, the credits are returned automatically. If it finishes later, you get it free. 3. You get the model you choose. For images and videos, the model you pick is the model that generates. GPT Image and Nano Banana render 2K / 4K natively, not by upscaling a smaller image. For chat, when one of a few Chinese models is temporarily unavailable, another model may answer instead; your usage log says so, and you pay the lower of the two prices. Good to know 4. Special tiers are cheaper but less stable. Video tiers marked "Special" are more likely to queue or fail at peak times. For important jobs, use the standard per-second tiers. 5. Render times vary. Times on model pages are recent measured medians; peak hours take longer. 6. Content review can reject a job. If the prompt, a reference file or the result fails review, the job fails and is not charged. We do not always receive the exact reason. 7. Results are kept for 15 days. Generated images and videos are deleted after 15 days, so please download them in time. If something fails Check the status page and announcements first. If the same request keeps failing, do not keep retrying right away: try again a little later, or switch to a standard per-second tier. When you contact us, include the task ID: support@kovapi.com, or our Telegram group (link at the bottom of this card). Published: 2026-10-10 07:51:44 UTC.
-
Notice
[Historical notice] Sunburst 2K reopened; Flare 2K / 4K remain under maintenance
The following records an earlier service state. Sunburst / Flare 2K is currently experiencing disruption; please pause retries. Latest 2K notice: /en/announcements#2026-10-18-sunburst-flare-2k-service-disruption The 2K tier of GPT Image 2.5 Sunburst has reopened for new website and API submissions, including text-to-image and image editing. The 2K and 4K tiers of GPT Image 2.5 Flare remain under maintenance; reopening will be announced separately. Requests rejected for maintenance are not charged; accepted jobs continue processing. The 1K tiers of both models, Sunburst 4K, and GPT Image 2 are outside this maintenance. The new prices for all three GPT image models are now effective: 5 credits per 2K image and 8 credits per 4K image; 1K is unchanged. Final charges continue to use the price at completion, including any difference due for jobs completed after the change. Price notice: /en/announcements#2026-10-10-gpt-image-2k-4k-price We apologize for the inconvenience. Published: 2026-10-10 06:07:56 UTC.
-
Price change
GPT image 2K / 4K price update: October 10, 12:00 UTC
Effective October 10, 2026 at 12:00 UTC, image generation prices for GPT Image 2, GPT Image 2.5 Flare, and GPT Image 2.5 Sunburst will change as follows: 2K: from 2 to 5 credits per image. 4K: from 5 to 8 credits per image. 1K prices remain unchanged. This applies to both text-to-image generation and image editing. The final charge uses the price in effect when generation completes. Requests submitted earlier but completed at or after the effective time will also use the new prices. Any difference from credits reserved at submission will be charged on completion. Failed generations are not charged. Thank you for your understanding and support. Published: 2026-10-10 05:27:41 UTC.
-
Notice
Image generation restored
Image generation models are now available again. We sincerely apologize for the inconvenience caused by the recent disruption. Thank you for your patience and understanding. Published: 2026-10-10 05:13:49 UTC.
-
Notice
[Previous notice] GPT image generation stability
This is a previous image generation status notice. For the current status, please see the latest announcement, “Image generation restored”. Published: 2026-10-10 03:18:33 UTC.
-
Notice
[Previous notice] Image model maintenance
This is a previous image generation status notice. For the current status, please see the latest announcement, “Image generation restored”. Published: 2026-10-10 01:12:20 UTC.
-
Notice
Video generation requests reopened
Video generation requests are open again on the website and API. We have added prompt checks before generation; requests blocked by these rules are not charged. Some models may still queue or fail to generate. Please check the model page for its status and avoid resubmitting tasks that are still processing. If you encounter an issue, contact support with the task ID. Thank you for your patience and understanding. Published: 2026-10-09 13:17:03 UTC.
-
Notice
[Ended] Temporary pause in video requests
This temporary pause has ended and video requests are open again. See the latest announcement for details. The original pause notice follows: We are experiencing issues with video generation and have temporarily paused all video models on the website and API. There is no confirmed recovery time. Please avoid repeated submissions; we will publish a notice when service resumes. New requests during the pause are not charged. Earlier tasks will be handled according to their results, with reserved credits refunded for failed tasks under the existing rules. We apologize for the inconvenience and thank you for your patience. Published: 2026-10-09 12:16:41 UTC.
-
Paused
Seedance 2.5 30s special-price capacity is being replenished
Capacity for the Seedance 2.5 30s special-price tier is exhausted. We're working to replenish it, with recovery expected Oct 9 at 12:00 UTC. Overload or generation-failed errors are expected while capacity is unavailable; failed tasks are not charged and reserved credits are refunded. Please use other available video models, such as Seedance 2.5 (per-second), Wan 3.0, or MiniMax H3, which are unaffected. Thank you for your patience. (Published Oct 9, 09:27 UTC.)
-
Notice
Generated images and videos are now kept for 15 days
To keep our cloud object-storage costs in check, we clean up files on a fixed schedule: generated images and videos are kept for 15 days (previously 60 days), and reference images / videos / audio you upload are kept for 7 days, after which they are deleted from object storage - please download anything you need before then. Call logs are kept for 60 days; the prompt and output text in them is cleared after 30 days, after which only the usage record (model, usage, time, cost) remains. Nothing is ever used for training or sold. (Posted Oct 7, 15:45 UTC / 23:45 Beijing time; updated Oct 9, 07:10 UTC / Oct 9, 15:10 Beijing time)
-
Notice
Wan: input-review failures and clearer error messages
Some recent Wan Standard and Prime tasks failed input review, including possible real human faces or rejected reference images. Wan is currently able to generate video, and a fresh test completed successfully. Our earlier errors did not clearly explain the restriction or what to change. We have corrected this: failed tasks return content_policy with the specific reason (FaceDetectionSuspect or DataInspectionFailed) and guidance. Replace the reference material instead of repeatedly submitting unchanged input; contact support if you suspect a false positive. Affected failed tasks were not charged and reserved credits were returned automatically. The API documentation now explains the face-review restriction in six languages. We are sorry for the unclear guidance. (Published: 2026-10-09 02:36:00 UTC / 2026-10-09 10:36:00 UTC+8.)
-
Notice
[Previous notice] Sunburst image model maintenance
This notice records an earlier incident. Sunburst 2K and 4K are currently unstable and under maintenance. Latest notice: /announcements#2026-10-13-sunburst-2k4k-maintenance This is a previous image generation status notice. For the current status, please see the latest announcement, “Image generation restored”. Published: 2026-10-08 22:57 UTC.
-
Notice
Seedance 2.5 submissions have resumed
A temporary issue caused some Seedance 2.5 video submissions to fail. Submissions have now resumed, and you can submit new tasks as usual. We are sorry for the disruption. The failed tasks affected by this incident were not charged, and reserved credits were returned automatically. If you still have trouble, please contact support@kovapi.com. We will keep improving the service. Thank you for your patience and support. (Published Oct 08, 22:47 UTC / 06:47 Beijing time.)
-
Notice
Our emails now come from our new domain, kovapi.com
Starting today, NezhaGate emails are moving to our new domain and will come from support@kovapi.com, including sign-up codes, security alerts and top-up notices. kovapi.com is our new domain: your account, balance, API keys and prices stay the same, and the current website and API addresses keep working. Add support@kovapi.com to your contacts so our emails don't end up in spam. We will never ask for your password or API key by email. (Published Oct 8, 12:30 UTC / 20:30 Beijing time)
-
Notice
Seedance 2.5 · 30s is back to normal; per-second Seedance 2.5 stays 10% off until Oct 9
Seedance 2.5 · 30s (seedance-2.5-30s) is back to normal and its success rate has recovered. It is still a fixed 30-second 720P clip for $1.50. A 30-second render currently takes about an hour, so submit and check back later, or pass a callback_url and we'll notify you when it's done. Failed jobs are never billed. The 10% discount on per-second Seedance 2.5 (seedance-2.5) continues until Oct 9, 24:00 Beijing time (16:00 UTC): 480P $0.09/s (was $0.10), 720P $0.142/s (was $0.158). Thank you for your patience these past days. (Posted Oct 6, 16:06 UTC / 00:06 Beijing time; updated Oct 8, 12:00 UTC / 20:00 Beijing time.)
-
Notice
Claude Opus 5.5 and Sonnet 5.5 are back
Claude Sonnet 5.5 is available again at the same price: $1 input, $5 output and $0.10 cached input per million tokens. Claude Opus 5.5 has been back since Oct 7, 11:37 UTC (19:37 Beijing time), also at the same price: $2 input, $10 output and $0.10 cached input. Both work on the OpenAI-compatible API and the native Anthropic API. (Posted Oct 7, 14:43 UTC / 22:43 Beijing time)
-
Price cut
Wan 3.0 Standard price cut: 720P now $0.065/s, 1080P now $0.13/s
Wan 3.0 Standard now costs $0.065 per second at 720P (was $0.08) and $0.13 per second at 1080P (was $0.16), both about 19% less. 480P stays at $0.04 per second, and Wan 3.0 Prime prices are unchanged. The new prices apply to tasks submitted from now on. (Posted Oct 7, 14:24 UTC / 22:24 Beijing time.)
-
New
Seedance 2.5 now in 1080P at $0.33 per second
Seedance 2.5 now renders in 1080P, billed at $0.33 per second for any length from 4 to 30 seconds (5 s is $1.65, 30 s is $9.90), with the same aspect ratios and reference options as 480P and 720P; renders take somewhat longer than at 720P. The API is unchanged: set resolution to 1080P. Failed jobs are refunded in full. 480P ($0.10 per second) and 720P ($0.158 per second) keep their prices. (Posted Oct 6, 09:54 UTC / 17:54 Beijing time.)
-
Price change
GPT Image 4K price change
Starting Oct 8, 00:00 Beijing time (Oct 7, 16:00 UTC), 4K images from GPT Image 2 and GPT Image 2.5 (Flare, Sunburst) cost 5 credits ($0.025) each instead of 4 credits ($0.02). 1K and 2K prices stay the same, and no other model is affected. (Posted Oct 6, 08:19 UTC / 16:19 Beijing time.)
-
Notice
Sign-up credits: 30 at sign-up, 50 more with your first top-up
From now on, new accounts get 30 free credits at sign-up and 50 more with their first top-up — 80 in total, as before. Only the first account registered from a network (IP address) within 7 days gets the 30 credits right away; later accounts from the same network receive both the 30 and the 50 with their first top-up, so nobody misses out. Existing accounts are not affected. Accounts mass-registered to farm credits will be banned. (Posted Oct 6, 03:35 UTC / 11:35 Beijing time.)
-
Notice
GPT Image is back to normal
Some GPT Image 2.5 (Flare / Sunburst) and GPT Image 2 requests failed or ran unusually long over the past two days. That's fixed: when the upstream hits a one-off error, we now retry automatically. Test generations are all succeeding again. Failed requests in that period were never billed. Thanks for your patience.
-
New
Seedance 2.5 · 30s at a special price: $1.50 per 30-second video
Seedance 2.5 · 30s is billed per clip: a 30-second 720P video is a flat $1.50 (about $0.05 a second), about 68% less than per-second Seedance 2.5 at 720P, in 16:9 or 9:16. It is in Beta while we tune render speed and success rate.
-
New
Seedance 2.0 · Clip at a special price: $1.00 per video
Seedance 2.0 · Clip is billed per clip: $1.00 whether it runs 5, 10 or 15 seconds. A 15-second video on per-second Seedance 2.0 (720P) costs $2.25, so this saves 56%. Output is 720P: 16:9 in 5 / 10 / 15 seconds, 9:16 at 15 seconds; up to 9 reference images, real-face references supported, native audio included. It is in Beta.
-
Notice
We're moving to a new domain: kovapi.com
We're starting to use our new domain, kovapi.com. Your account, balance, API keys and prices stay the same, and the current API address (nezhagate.com) will keep working, so there's no code to change. We'll announce the switch date here and in our Telegram group ahead of time.
-
Price change
New prices for six GPT models from October 5
From 10:00 Beijing time on October 5 (02:00 UTC): GPT-6 Astra $3 / $15, GPT-5.5 $1.5 / $9, GPT-5.6 Sol $1.2 / $6, GPT-5.6 Terra $0.6 / $3.6, GPT-6.1 Sol and GPT-6 Sol $0.6 / $3 per 1M input / output tokens, still about 70% below the official list price. Both Luna models keep their prices.
-
New
Claude Opus 5.5 and Sonnet 5.5 are live
Claude Opus 5.5 ($2 / $10) and Claude Sonnet 5.5 ($1 / $5) are now available, with cache reads at $0.10 (all per 1M tokens). Both work through the OpenAI-compatible API and the native Anthropic API; effort, thinking summaries and structured output need the native /anthropic endpoint.
-
New
Grok Imagine image generation and editing are live
Standard costs $0.005 and Quality $0.01 per image; 1K and 2K cost the same, and 4K is not offered. Editing costs the same as generation.
-
Price change
New prices for DeepSeek and GLM-5.3
From 00:00 Beijing time on October 4 (16:00 UTC, October 3): DeepSeek V4.1 Flash and DeepSeek V4 Flash 0731 cost $0.13 / $0.51 (cache reads $0.004) and GLM-5.3 costs $1.12 / $3.52 (cache reads $0.21), all per 1M tokens and still below the official list price.
-
Price cut
Seedance 2.5 at 480P drops to $0.10 per second
Seedance 2.5 at 480P now costs $0.10 per second; 720P stays at $0.158 per second.
-
New
Veo 3.1 and Gemini Omni Flash are live
Veo 3.1 opens its Fast tier: $0.20 per 8-second clip at 720p and $0.22 at 1080p. Gemini Omni Flash costs $0.05 per second, in 6, 8 or 10-second clips.
-
Paused
Seedance 1080P is paused
1080P on Seedance 2.5 and Seedance 2.0 did not meet our success-rate bar, so it is paused; 480P and 720P are unaffected. We will post here when it is back.
For scripts: /announcements.json · Service status →
