Compare AI models
| Specifications | ||
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 1.05M | 1.05M |
| API pricingUSD per 1M tokens · lower wins · base tier | ||
Input priceWhat you pay for everything you send the model — your question, plus any documents or earlier conversation you include with it. | $8.00 | $2.00 |
Output priceWhat you pay for the text the model writes back. It is normally the dearer half: producing an answer costs more than reading one. | $40.00 | $10.00 |
Cached input priceA reduced rate for text you send over and over. If every request starts with the same instructions or the same document, the provider keeps a copy ready and charges less to read it again. | $0.40 | $0.10 |
| $3,600OpenAI list price | $900OpenAI list price | |
| Benchmarks | ||
Blocker Deception (OpenAI) · without Codex promptBlocker disclosure — share of coding trials where the model omitted or misrepresented an obstacle without the Codex developer prompt | 22.28% | 24.08% |
Blocker Deception (OpenAI) · with Codex promptBlocker disclosure — share of coding trials where the model omitted or misrepresented an obstacle with the Codex developer prompt | 8.04% | 8.88% |
Simulated Workplace Tasks (OpenAI)High-severity improper actions — share of scored simulated workplace tasks with a high-severity improper action | 0.64% | 0.71% |
FrontierCode 2.0Agentic coding — 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 | 44.2% | 43.8% |
BenchCAD 2.1 CoreCAD programming — partial-credit score for modeling hardware designs in the held-out Core evaluation | 82.5% | 73.1% |
SecureSWESecure code generation — OpenAI’s internal evaluation of writing code that is both functional and secure | 65.7% | 55.6% |
PatchBenchSecurity patching — security patches that fix vulnerabilities while preserving intended behavior | 79.7% | 75.6% |
Terminal-Bench-Science 0.1Agentic scientific computing — 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. | 77.1% | 68% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 97% | 96% |
Vals Science Index v1Scientific reasoning — Vals AI’s composite accuracy across three scientific reasoning evaluations | 68.4% | 58.5% |
Agents’ Last Exam v1.1Agentic computer use — averaged rubric score on the 122-task v1.1 suite of long-horizon professional workflows | 78.3% | 73% |
AutomationBench 1.0.6Business workflows — end-to-end business workflow automation on the version 1.0.6 task set | 52.4% | 41.1% |
LifeValEveryday tasks — OpenAI’s rubric score for completing practical everyday tasks tailored to a person’s needs | 86.1% | 83.8% |
Harvey LAB-AA v1.1Agentic legal work — full-credit rate on legal assignments, requiring every grading criterion to be met without material hallucinations | 15% | 11.1% |
HealthBench Professional · OpenAI grading, rawHealth — unadjusted physician-rubric score using the official OpenAI grading procedure | 66% | 65.5% |
HealthBench Professional · OpenAI grading, length-adjustedHealth — official OpenAI grading with the response-length penalty applied | 66.3% | 65.9% |
MentalHealthBenchMental health — overall clinician-rubric score on mental health conversations | 53.3% | 50.9% |
AA Intelligence Index v5.0.1Overall intelligence — Artificial Analysis’s version 5.0.1 composite across nine evaluations of knowledge work, coding, science, and general knowledge | 62.1 | 57.3 |
GDPval v2 (OpenAI Internal)Knowledge work — OpenAI’s internal score on multistep professional assignments drawn from consulting, finance, and design projects | 50.4% | 46.9% |
AA-3DCommunity preference (interactive 3D) — blind human preferences between paired interactive 3D experiences built from the same prompt | 1318 | 1259 |
AA-AnalystAgent (pass^5)Financial research — share of quantitative questions about real spreadsheets and reports answered correctly on all five independent attempts | 62.5% | — |
| Overview | ||
| Company | OpenAI | OpenAI |
| Release date | Nov 17 2026 | Nov 17 2026 |
| Access | Closed | Closed |
| Model details | View model | View model |
Other comparisons
GPT-6.2 AstravsClaude Fable 5.2GPT-6.2 SolvsClaude Fable 5.2GPT-6.2 AstravsGemini 4 HydrogenGPT-6.2 SolvsGemini 4 HydrogenGPT-6.2 AstravsMuse EmberGPT-6.2 SolvsMuse EmberGPT-6.2 AstravsGrok 4.7GPT-6.2 SolvsGrok 4.7GPT-6.2 AstravsDeepSeek-V4.1-Flash-1017GPT-6.2 SolvsDeepSeek-V4.1-Flash-1017GPT-6.2 AstravsMistral Large 4GPT-6.2 SolvsMistral Large 4Frequently asked questions
GPT-6.2 Astra leads GPT-6.2 Sol on 20 of the 20 benchmarks they both report. GPT-6.2 Sol is cheaper on both input and output: $2.00 vs $8.00 per million input tokens, and $10.00 vs $40.00 per million output tokens. Figures are base-tier rates. Both models were released on the same day — Nov 17 2026.
Context windows are 1.05M (GPT-6.2 Astra) vs 1.05M (GPT-6.2 Sol).
On Blocker Deception (OpenAI) · without Codex prompt, GPT-6.2 Astra leads at 22.28% vs GPT-6.2 Sol at 24.08%. On Blocker Deception (OpenAI) · with Codex prompt, GPT-6.2 Astra leads at 8.04% vs GPT-6.2 Sol at 8.88%. On Simulated Workplace Tasks (OpenAI), GPT-6.2 Astra leads at 0.64% vs GPT-6.2 Sol at 0.71%. On FrontierCode 2.0, GPT-6.2 Astra leads at 44.2% vs GPT-6.2 Sol at 43.8%. On BenchCAD 2.1 Core, GPT-6.2 Astra leads at 82.5% vs GPT-6.2 Sol at 73.1%. On SecureSWE, GPT-6.2 Astra leads at 65.7% vs GPT-6.2 Sol at 55.6%. On PatchBench, GPT-6.2 Astra leads at 79.7% vs GPT-6.2 Sol at 75.6%. On Terminal-Bench-Science 0.1, GPT-6.2 Astra leads at 77.1% vs GPT-6.2 Sol at 68%. On GPQA Diamond, GPT-6.2 Astra leads at 97% vs GPT-6.2 Sol at 96%. On Vals Science Index v1, GPT-6.2 Astra leads at 68.4% vs GPT-6.2 Sol at 58.5%. On Agents’ Last Exam v1.1, GPT-6.2 Astra leads at 78.3% vs GPT-6.2 Sol at 73%. On AutomationBench 1.0.6, GPT-6.2 Astra leads at 52.4% vs GPT-6.2 Sol at 41.1%. On LifeVal, GPT-6.2 Astra leads at 86.1% vs GPT-6.2 Sol at 83.8%. On Harvey LAB-AA v1.1, GPT-6.2 Astra leads at 15% vs GPT-6.2 Sol at 11.1%. On HealthBench Professional · OpenAI grading, raw, GPT-6.2 Astra leads at 66% vs GPT-6.2 Sol at 65.5%. On HealthBench Professional · OpenAI grading, length-adjusted, GPT-6.2 Astra leads at 66.3% vs GPT-6.2 Sol at 65.9%. On MentalHealthBench, GPT-6.2 Astra leads at 53.3% vs GPT-6.2 Sol at 50.9%. On AA Intelligence Index v5.0.1, GPT-6.2 Astra leads at 62.1 vs GPT-6.2 Sol at 57.3. On GDPval v2 (OpenAI Internal), GPT-6.2 Astra leads at 50.4% vs GPT-6.2 Sol at 46.9%. On AA-3D, GPT-6.2 Astra leads at 1318 vs GPT-6.2 Sol at 1259.