# GPT-6.2 Astra and Sol

> GPT-6.2 has two models, Astra and Sol. Model IDs, 1.05M context, prices, safety ratings and which one to pick for agents, as of 21 November 2026.

Source: https://www.honeyhive.ai/learn/gpt-6-2 (updated 2026-11-21)

Start with GPT-6.2 Sol for most agent and coding work and reserve Astra for tasks where your own tests show Sol falls short. OpenAI [describes](https://openai.com/index/introducing-gpt-6-2-astra-sol/) Astra as its frontier model for the hardest projects and Sol as the efficient, versatile model for everyday work. Both are in the API, and in Codex and ChatGPT Work for Plus, Pro, Enterprise and Edu users.

Prices and limits below were read from OpenAI's pages on 21 November 2026 and change often.

## The two models side by side

| | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| API model ID | [`gpt-6.2-astra`](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [`gpt-6.2-sol`](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Role | Hardest projects | Everyday work |
| Context window | [1,050,000 tokens](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [1,050,000 tokens](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Max input / output | [922,000 / 128,000 tokens](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [922,000 / 128,000 tokens](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Input per 1M tokens | [$8](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [$2](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Cached input per 1M | [$0.40](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [$0.10](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Cache writes per 1M (30-minute retention) | [$10](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [$2.50](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Output per 1M tokens | [$40](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [$10](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |
| Knowledge cutoff | [May 18, 2026](https://developers.openai.com/api/docs/models/gpt-6.2-astra) | [June 1, 2026](https://developers.openai.com/api/docs/models/gpt-6.2-sol) |

Both model pages list support for streaming, structured outputs, function calling, file search, image input, web search and prompt caching. Both pages ([Astra](https://developers.openai.com/api/docs/models/gpt-6.2-astra), [Sol](https://developers.openai.com/api/docs/models/gpt-6.2-sol)) list fine-tuning and predicted outputs as unsupported, and both say the API rejects requests that set reasoning effort to `none`; see [reasoning effort](/learn/reasoning-effort) for how the setting changes cost and quality.

## What changed in pricing

OpenAI [says](https://openai.com/index/introducing-gpt-6-2-astra-sol/) Astra's input and output prices are 20% lower than GPT-6 Astra's, with cached input at a 60% discount, and that Sol keeps GPT-6.1 Sol's prices. An "Ultrafast" speed mode is available for both models in the API.

## What OpenAI claims, and what to check

OpenAI's launch post [reports](https://openai.com/index/introducing-gpt-6-2-astra-sol/) Sol coming within 0.5 points of Astra on FrontierCode 2.0 at about one-fifth the cost, with both run at max reasoning effort, and Astra scoring 5.2 points above GPT-6 Astra on [Agents Last Exam](/learn/agents-last-exam) at less than a third of the cost. These are the vendor's numbers. A max-effort result tells you little about the cost of your own workload, because thinking tokens are billed. Run the same task set through Sol and Astra and compare accuracy, cost per finished task and latency from traces. See [LLM cost optimization](/learn/llm-cost-optimization) and [model routing](/learn/model-routing) for sending easy work to the cheaper model.

## Safety ratings

OpenAI's launch post [rates](https://openai.com/index/introducing-gpt-6-2-astra-sol/) both models Critical for cybersecurity and High for biological and chemical risk under its Preparedness Framework. It rates Astra High for AI self-improvement and calls it the first externally released model at that level. OpenAI [launched](https://openai.com/index/introducing-gpt-6-2-astra-sol/) with tighter safeguards on higher-risk cybersecurity help, which also restrict some legitimate security testing, and says it will bring the GPT-6.2 models to Daybreak, its access program for defenders.

For agent builders the alignment note matters most. OpenAI [reports](https://openai.com/index/introducing-gpt-6-2-astra-sol/) that in two tests the models were more likely than GPT-6 Astra to omit obstacles that blocked their work, that it added Codex instructions to reduce this, and that it is still investigating. An agent that hides a failed step is a reason to trace tool calls and check final claims against them. See [agent tracing](/learn/agent-tracing) and [AI agent guardrails](/learn/ai-agent-guardrails).

## FAQ

### What are the GPT-6.2 model IDs?

The API IDs are `gpt-6.2-astra` and `gpt-6.2-sol`, according to OpenAI's [launch post](https://openai.com/index/introducing-gpt-6-2-astra-sol/). Astra is the frontier model for the hardest projects and Sol is the efficient model for everyday work.

### How much does GPT-6.2 cost?

As of 21 November 2026, OpenAI's model pages list Astra at [$8 input and $40 output](https://developers.openai.com/api/docs/models/gpt-6.2-astra) per 1M tokens and Sol at [$2 input and $10 output](https://developers.openai.com/api/docs/models/gpt-6.2-sol). Cached input is $0.40 for Astra and $0.10 for Sol.

### What is the GPT-6.2 context window?

Both models list a [1,050,000-token context window](https://developers.openai.com/api/docs/models/gpt-6.2-astra), a 922,000-token maximum input and 128,000 maximum output tokens. See [context window](/learn/context-window) for what the limit means in practice.

### Should I use GPT-6.2 Astra or Sol?

Start with Sol and move a task to Astra only if your own evaluation shows a gap. OpenAI says Sol is [within 0.5 points](https://openai.com/index/introducing-gpt-6-2-astra-sol/) of Astra on FrontierCode 2.0 at about one-fifth the cost, a vendor result you should confirm on your tasks.
