KovAPIKovAPI
Chat ● Operational Service status →

GPT-6.2 Astra

⧉
gpt-6.2-astra
Model Type Astra6.2 Astra6.2 Sol6.1 SolSolLuna

GPT-6.2 Astra is the new flagship OpenAI released on November 17, 2026, listed 20% below GPT-6 Astra ($8 input and $40 output per million tokens). It has a 1,050,000-token context, up to 128,000 output tokens per call, reasoning_effort in five steps (low / medium / high / xhigh / max), image input, tool calling and structured output. Fully OpenAI-SDK compatible: set model to gpt-6.2-astra on /chat/completions or /responses. Pay-as-you-go, failed calls are never billed; cached input bills at one twentieth of the input rate, and we do not add the official surcharge for prompts above 272,000 tokens.

ChatGPT-6 familyFlagship1.05M contextPrompt caching

Live Test · Playground

Try out GPT-6.2 Astra right here (available after login).

Input

Advanced

The upstream for this model does not accept temperature / top_p / max_tokens (they are silently ignored). Use reasoning_effort below for thinking depth, and prompt wording for length.

Web search
Memory · multi-turn

Conversation

Start a conversationType a message below to begin

About GPT-6.2 Astra

GPT-6.2 Astra is the new flagship OpenAI released on November 17, 2026, served on KovAPI through the OpenAI-compatible API. Its list price is 20% below GPT-6 Astra: $8 input and $40 output per million tokens. It has a 1,050,000-token context, up to 128,000 output tokens per call, reasoning_effort in five steps (low / medium / high / xhigh / max), image input, tool calling and structured output, on both /chat/completions and /responses. Set model to gpt-6.2-astra and keep your SDK and request shape. Pay-as-you-go, failed calls are never billed, and cached input settles at one twentieth of the input rate.

Use cases

The hardest reasoning and research work

Multi-step math and logic, reviews of complex designs, long analytical reports: give work that needs deep thinking to Astra and raise effort to xhigh or max when needed.

Agentic coding

Multi-file refactors, cross-module debugging, whole features from a spec. Codex, Cursor and similar tools speak the OpenAI-compatible endpoint, so switching is a one-field change.

Long documents and codebases

1,050,000 tokens hold a full set of contracts, financial reports or a whole repository in one call, with no chunking and stitching of your own.

Tool calling and structured output

Function calls come back in the OpenAI format and strict json_schema output can go straight into your database, which suits decision steps inside automated pipelines.

How to choose

Pick GPT-6.2 Astra when you need the strongest reasoning and coding. GPT-6.2 Sol, from the same release, comes close at a quarter of the price and is the better buy for everyday coding and agent work; use GPT-6 Luna for simple high-volume, low-latency jobs. They all belong to the GPT-6 family and share a 1,050,000-token context, so switching is a one-field change.

FAQ

How is it different from GPT-6 Astra?
GPT-6.2 Astra is OpenAI's November flagship, listed 20% below GPT-6 Astra ($8 / $40 versus $10 / $50) with the same context and output limits. On KovAPI it is also cheaper than gpt-6-astra.
Which reasoning_effort levels are supported?
Five: low / medium / high / xhigh / max. The upstream rejects none and minimal, so the gateway maps both to low. When omitted the gateway requests medium, the model's own default, and returns the thinking summary; more thinking means more output tokens, billed at actual usage.
What are the context and output limits?
A 1,050,000-token context and up to 128,000 output tokens per call. The official API charges more for prompts above 272,000 tokens; we do not.
How is caching billed?
Cached input settles at one twentieth of the input rate ($0.12 per million tokens). The upstream decides what is cached; long repeated prefixes hit more often, but a hit is not guaranteed every time.
Can I use /v1/responses?
Yes. The same model name works on /v1/chat/completions and /v1/responses, and both support streaming and tool calling.
Am I charged for failed calls?
No. Only calls that return a proper response are settled at actual usage; upstream errors and timeouts never touch your balance.

Related Models

Explore other models you can integrate.

View all →