# GPT-6.2 Sol

GPT-6.2 Sol is an AI model released by OpenAI on Nov 17 2026. It has a 1.05M context window. Tracked results include 96% on GPQA Diamond, 43.8% on FrontierCode 2.0 and 73.1% on BenchCAD 2.1 Core.

## Facts

| Field | Value |
| --- | --- |
| Model | GPT-6.2 Sol |
| Developer | OpenAI |
| Release date | Tuesday, Nov 17 2026 |
| Licensing | Closed |
| Context window | 1.05M |

## API pricing

All rates in USD per 1,000,000 tokens, pay-as-you-go.

| Tier | Input | Cached input | Cache write | Output |
| --- | --- | --- | --- | --- |
| Up to 272K input tokens | $2.00 | $0.10 | $2.50 | $10.00 |
| Over 272K input tokens | $4.00 | $0.20 | $5.00 | $15.00 |

OpenAI also publishes a Fast mode for this model at exactly twice the standard rate in every column — $4.00 per million input tokens and $20.00 per million output at short context.
At standard rates, 12-hour cache writes cost $5 per million tokens up to 272K input tokens and $10 above that threshold.
Ultrafast costs 3× the standard rate through February 28, 2027 and 6× starting March 1, 2027 (Pacific time).
Eligible Responses API requests in the background completion beta receive 30% off standard model token rates with a 2-hour window or 40% off with an 8-hour window.
Verified November 17, 2026 against the first-party source: https://developers.openai.com/api/docs/pricing

## Tracked benchmark scores

| Benchmark | Score | Source | What it measures |
| --- | --- | --- | --- |
| Blocker Deception (OpenAI) (without Codex prompt) | 24.08% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | share of coding trials where the model omitted or misrepresented an obstacle without the Codex developer prompt |
| Blocker Deception (OpenAI) (with Codex prompt) | 8.88% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | share of coding trials where the model omitted or misrepresented an obstacle with the Codex developer prompt |
| Simulated Workplace Tasks (OpenAI) | 0.71% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | share of scored simulated workplace tasks with a high-severity improper action |
| FrontierCode 2.0 | 43.8% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | whether a model’s code changes to open-source repositories would be merged by their maintainers (version 2.0); separate from FrontierCode v1.1 scores |
| BenchCAD 2.1 Core | 73.1% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | partial-credit score for modeling hardware designs in the held-out Core evaluation |
| SecureSWE | 55.6% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | OpenAI’s internal evaluation of writing code that is both functional and secure |
| PatchBench | 75.6% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | security patches that fix vulnerabilities while preserving intended behavior |
| Terminal-Bench-Science 0.1 | 68% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | The same command-line setup as Terminal-Bench, pointed at scientific work: the AI has to drive research tooling and computational workflows through to a result, rather than administer a machine. Version 0.1 is the first release of the task set, and scores run lower than on the general board. Higher is better. |
| GPQA Diamond | 96% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. |
| Vals Science Index v1 | 58.5% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | Vals AI’s composite accuracy across three scientific reasoning evaluations |
| Agents’ Last Exam v1.1 | 73% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | averaged rubric score on the 122-task v1.1 suite of long-horizon professional workflows |
| AutomationBench 1.0.6 | 41.1% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | end-to-end business workflow automation on the version 1.0.6 task set |
| LifeVal | 83.8% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | OpenAI’s rubric score for completing practical everyday tasks tailored to a person’s needs |
| Harvey LAB-AA v1.1 | 11.1% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | full-credit rate on legal assignments, requiring every grading criterion to be met without material hallucinations |
| HealthBench Professional (OpenAI grading, raw) | 65.5% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | unadjusted physician-rubric score using the official OpenAI grading procedure |
| HealthBench Professional (OpenAI grading, length-adjusted) | 65.9% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | official OpenAI grading with the response-length penalty applied |
| MentalHealthBench | 50.9% | [OpenAI GPT-6.2 system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol), retrieved 2026-11-17 | overall clinician-rubric score on mental health conversations |
| AA Intelligence Index v5.0.1 | 57.3 | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | Artificial Analysis’s version 5.0.1 composite across nine evaluations of knowledge work, coding, science, and general knowledge |
| GDPval v2 (OpenAI Internal) | 46.9% | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | OpenAI’s internal score on multistep professional assignments drawn from consulting, finance, and design projects |
| AA-3D | 1259 | [OpenAI GPT-6.2 launch](https://openai.com/index/introducing-gpt-6-2-astra-sol/), retrieved 2026-11-17 | blind human preferences between paired interactive 3D experiences built from the same prompt |

## About GPT-6.2 Sol

GPT-6.2 Sol was introduced on November 17, 2026 with GPT-6.2 Astra as the lower-cost model for everyday coding and agent work. OpenAI described it as smarter and more token-efficient than GPT-6.1 Sol, and said many of its own employees used it as their default. It reached Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex the same day, and the API as gpt-6.2-sol. Pro 500 subscribers could use Ultrafast on Sol, and Pro 200 subscribers received Ultrafast on Sol at launch, which OpenAI described as the first time that tier was offered to Pro 200.

The API accepts text and images and returns text. It has a 1,050,000-token context window, with limits of 922,000 input and 128,000 output tokens. Its default context in Codex is 500,000 tokens. The knowledge cutoff is June 1, 2026, and reasoning effort ranges from low through maximum. Both Responses and Chat Completions are available, with tool calling through Responses. Standard short-context rates stayed at GPT-6.1 Sol's $2 per million input tokens, $0.10 cached and $10 output.

OpenAI reported 43.8% on Cognition's FrontierCode 2.0 at maximum effort, close to GPT-6.2 Astra at about one-fifth the cost. On LifeVal, Sol scored 83.8% and outscored GPT-6 Astra at about one-seventh the cost. On OpenAI's internal GDPval v2, Sol improved 5.3 points on GPT-6.1 Sol. OpenAI rated Sol Critical for cybersecurity and High for biological and chemical capabilities. The system card places Sol below High for AI self-improvement.

## Questions and answers

### When was GPT-6.2 Sol released?

GPT-6.2 Sol was released by OpenAI on Tuesday, Nov 17 2026.

### Who made GPT-6.2 Sol?

GPT-6.2 Sol was built by OpenAI. Creators of ChatGPT and the GPT series of models. Pioneered large-scale language model research.

### How much does GPT-6.2 Sol cost?

GPT-6.2 Sol costs $2.00 per million input tokens and $10.00 per million output tokens through the OpenAI API. Cached input is $0.10 per million tokens. Those are the rates for the “Up to 272K input tokens” tier; 1 other pricing tier is published for this model. Rates are pay-as-you-go API prices verified against OpenAI's published pricing on November 17, 2026.

### What benchmark scores did GPT-6.2 Sol get?

GPT-6.2 Sol reports 20 tracked benchmark scores — Blocker Deception (OpenAI) (without Codex prompt): 24.08%; Blocker Deception (OpenAI) (with Codex prompt): 8.88%; Simulated Workplace Tasks (OpenAI): 0.71%; FrontierCode 2.0: 43.8%; BenchCAD 2.1 Core: 73.1%; SecureSWE: 55.6%; PatchBench: 75.6%; Terminal-Bench-Science 0.1: 68%; GPQA Diamond: 96%; Vals Science Index v1: 58.5%; Agents’ Last Exam v1.1: 73%; AutomationBench 1.0.6: 41.1%; LifeVal: 83.8%; Harvey LAB-AA v1.1: 11.1%; HealthBench Professional (OpenAI grading, raw): 65.5%; HealthBench Professional (OpenAI grading, length-adjusted): 65.9%; MentalHealthBench: 50.9%; AA Intelligence Index v5.0.1: 57.3; GDPval v2 (OpenAI Internal): 46.9%; AA-3D: 1259. Tracked scores may come from lab reports or independent benchmarks; source details accompany the benchmark data.

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

GPT-6.2 Sol has a context window of 1.05M. Providers may specify separate input and output token limits.

### Is GPT-6.2 Sol open source?

No. GPT-6.2 Sol is a closed model. The weights are not published — it is available only through the provider's own API, apps, or partner platforms.

### What came before and after GPT-6.2 Sol?

OpenAI's previous tracked release was GPT-6.2 Astra on Nov 17 2026. It is the most recent OpenAI model tracked on AI Release Tracker.


---

Canonical page: https://aireleasetracker.com/model/openai/gpt-6.2-sol
Site index: https://aireleasetracker.com/llms.txt
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.
