All vikis
007 First Light
AI
Aion 2
Ananta
ARC Raiders
ArcheAge Chronicles
Call of Duty: Modern Warfare 4
Chrono Odyssey
Civilization 3
Civilization 4
Claude
Crimson Desert
Deadlock
Diablo 4
Diablo 5
Enshrouded
Fable
Final Fantasy Resonance
Football Manager 27
Forza Horizon 6
God of War
Grand Theft Auto 6
Heroes of Might and Magic 3OpenAI Benchmarks
As of Nov 6, 2026, no OpenAI model leads the main independent leaderboards: GPT-6 Astra scores 57 on Artificial Analysis's index behind Gemini 4 Carbon (60) and Claude Opus 5.5 (59), ranks 2nd on LMArena WebDev and 37th on LMArena Text, while OpenAI's own tables show it ahead on many agent, math and cyber tests.
On the independent leaderboards read on Nov 6, 2026, OpenAI's best model is near the top but not first: GPT-6 Astra (max) scores 57 on the Artificial Analysis Intelligence Index, behind Gemini 4 Carbon (60) and Claude Opus 5.5 (59), ranks 2nd on LMArena's WebDev board behind Claude Opus 5.5, and sits 37th on LMArena's text board, where OpenAI's best is GPT-5.6 Sol at 22nd. OpenAI's own launch tables tell a different story, with GPT-6 Astra ahead on most agent, math and security tests it chose. Both are real numbers measuring different things; this page sets them side by side with their dates.
Quick facts
| Artificial Analysis Intelligence Index (v5) | 1 Gemini 4 Carbon 60, 2 Claude Opus 5.5 59, GPT-6 Astra 57, read Nov 6, 2026 |
| LMArena Text | 1 Gemini 4 Carbon 1528; best OpenAI GPT-5.6 Sol, 22nd, 1485, updated Oct 27, 2026 |
| LMArena WebDev | 1 Claude Opus 5.5 1810, 2 GPT-6 Astra 1773, updated Oct 27, 2026 |
| LMArena Vision | 1 Claude Fable 5 1309; best OpenAI GPT-6.1 Sol, 9th, 1294, updated Oct 19, 2026 |
| OpenAI's headline claims | FrontierMath Tier 4 97.6%, ARC-AGI-3 99.9% for GPT-6 Astra (Sep 3, 2026, OpenAI's own) |
| SWE-bench | No current GPT-6 result published |
Independent leaderboards
Artificial Analysis Intelligence Index. Artificial Analysis runs the same nine tests on every model itself (agent tasks, coding in a terminal, science, knowledge and others) and combines them into one score. Version 5 was announced Oct 27, 2026, with methodology v5.0.1 on Nov 3.
| Rank | Model | Maker | Score | Cost per index task |
|---|---|---|---|---|
| 1 | Gemini 4 Carbon (high) | 60 | $7.76 | |
| 2 | Claude Opus 5.5 (max) | Anthropic | 59 | $10.04 |
| 5 | GPT-6 Astra (max) | OpenAI | 57 | $6.24 |
| 8 | GPT-6.1 Sol (max) | OpenAI | 55 | $1.27 |
| 9 | Claude Sonnet 5.5 (max) | Anthropic | 55 | not recorded |
| 14 | GPT-6.1 Terra (max) | OpenAI | 51 | $0.96 |
| 18 | Claude Fable 5.1 (xhigh) | Anthropic | 50 | not recorded |
| tied 35 | Grok 4.7 (xhigh) | xAI | 35 | not recorded |
| 42 | DeepSeek V4.1 Flash (max) | DeepSeek | 33 | not recorded |
| GPT-6 Luna (max) | OpenAI | 28 | $0.15 |
Source: Artificial Analysis model leaderboard, read Nov 6, 2026 (the table shows no date of its own).
LMArena (now Arena, at arena.ai). People compare two anonymous answers and vote; the ratings are Elo-style scores from millions of votes, so they measure what people prefer, not whether answers are correct.
| Board (date) | Leader | Best OpenAI | Other OpenAI models |
|---|---|---|---|
| Text (Oct 27, 2026, 9.0M votes) | Gemini 4 Carbon (High) 1528 | 22nd GPT-5.6 Sol (xHigh) 1485 | 27th GPT-6.1 Sol 1481; 37th GPT-6 Astra 1476; 97th GPT-6 Luna 1441 |
| WebDev (Oct 27, 2026) | Claude Opus 5.5 (Max) 1810 | 2nd GPT-6 Astra (Max) 1773 | 5th GPT-6.1 Sol 1751; 14th GPT-6.1 Terra 1673 |
| Vision (Oct 19, 2026) | Claude Fable 5 (High) 1309 | 9th GPT-6.1 Sol (Max) 1294 | GPT-6 Astra not in the top 20 |
| Agents (read Nov 6, 2026) | Claude Opus 5.5 (Max) 20.14% | 4th GPT-6 Astra (Max) 14.70% | 6th GPT-6.1 Sol 12.51%; 7th GPT-6.1 Terra 11.40% |
On the text board, rival labs' best models sit close to GPT-6 Astra (1476): DeepSeek-V4.1-Flash is 35th at 1476 and xAI's Grok 4.20 Beta 1 (listed as SpaceXAI) 40th at 1475.
SWE-bench and others. The public SWE-bench Pro board from Scale AI looks stale: its newest OpenAI entry is gpt-5.4 (xHigh) at 59.10, second behind Meta's Muse Spark 1.1, with no GPT-6 model and no dates. SWE-bench Verified's published data has no entry after Feb 26, 2026. The Terminal-Bench 4.0 and ARC Prize tables could not be read as text on Nov 6, 2026; ARC Prize says its 2026 results come on Dec 4, 2026. No reliable independent coding-benchmark comparison of GPT-6 exists yet beyond Arena WebDev.
OpenAI's own published scores
OpenAI's GPT-6 Astra announcement compares it with GPT-5.6 Sol, three Claude models and Gemini 3.8 Flash, each at its best effort. A selection, in percent:
| Benchmark | GPT-6 Astra | GPT-5.6 Sol | Claude Fable 5.1 | Notes |
|---|---|---|---|---|
| Terminal-Bench 4.0 | 57.9 | 37.3 | 55.8 | Coding in a terminal |
| AutomationBench | 41.4 | 18.1 | 31.4 | Business automation |
| GPQA Diamond | 96.0 | 94.6 | 93.7 | Graduate science |
| Humanity's Last Exam (tools) | 57.2 | not reported | 65.0 | Claude ahead |
| HealthBench Professional | 63.4 | 60.5 | 58.1 | 64.7 in the corrected system card |
Source: Introducing GPT-6 Astra, OpenAI, Sep 3, 2026; HealthBench correction from the GPT-6 Astra System Card.
For the smaller models, OpenAI reported that GPT-6.1 Sol came within 2.1 points of Astra on OSWorld 2.0 (offline) at about one-seventh of the cost per task, and that GPT-6.1 Terra scored 50.3% on Terminal-Bench 4.0 against 21.5% for GPT-5.6 Terra. OpenAI publishes no SWE-bench Verified or SWE-bench Pro score for Astra.
What rivals publish about GPT-6
Anthropic's Claude Opus 5.5 launch (Sep 22, 2026) included GPT-6 Astra in its own table. There Opus 5.5 leads Astra on Terminal-Bench 4.0 (66.4% to 57.9%) and Humanity's Last Exam with tools (67.7% to 57.2%), while Astra leads on AutomationBench (41.4% to 40.0%) and Terminal-Bench Science (64.6% to 58.7%). Like OpenAI's figures, these are the maker's own. Google's Gemini 4 Argon announcement names no GPT or Claude model.
How to read these numbers
- Check the date and version. OpenAI's Astra table quotes Artificial Analysis's older v4.1.1 index, where Astra scored 61.2 and Claude Fable 5.1 65.7; on today's v5 the same models score 57 and 50. Neither is wrong; they are different tests.
- Makers choose their tests. OpenAI's and Anthropic's tables each show their model winning most rows. Independent boards, which run every model the same way, are the fairer comparison.
- Effort level matters. Every score above names a setting (max, xhigh, high). In ChatGPT, Plus does not include Extra High or Pro, so a Plus user is not getting the "max" numbers. See ChatGPT Models.
- Cost is part of the result. On Artificial Analysis, GPT-6.1 Sol reaches 55 for $1.27 a task, against $10.04 for Claude Opus 5.5's 59.
- Preference is not accuracy. LMArena measures which answer people liked better. It is a good guide to style and helpfulness, a weaker one to correctness.
- Some scores are contested. OpenAI's system card notes possible contamination on ExploitBench, where Astra scored 100%.
For product-level comparisons, see ChatGPT vs Claude and ChatGPT vs Gemini.
What testers found
In an Aug 27, 2026 video comparing Codex with Claude Code, Tim Ruscica of Tech With Tim showed how OpenAI's GPT-5.6 launch chart and Anthropic's Claude Opus 5 chart each led with different benchmarks that favored their own model, and called it "marking their own homework". On an independent coding benchmark refreshed at the end of July 2026 that also grades code quality, he showed Claude Fable 5, Claude Opus 5 and GPT-5.6 tied at 34.7%, on about 100 tasks.
Matt Wolfe, reading OpenAI's GPT-6 Astra table in a video recorded on Sep 3, 2026, singled out the jump on AutomationBench but said Astra's 74.1% on DeepSWE, about 2 points over GPT-5.6 Sol, was lower than he expected, since Gemini 3.8 Flash and Claude Opus 5 sat at about 74% and Meta's own site gave Muse Spark 1.3 75.4%.
Fireship's Sep 4, 2026 Code Report made the same point about the index: at launch Artificial Analysis gave Astra 61, "exactly the same" as GPT-5.6 Sol and 5 points behind Claude Fable 5.1, which fits the v4.1.1 figures OpenAI itself quoted.
Frequently asked questions
Is GPT-6 the best AI model? Not on the independent leaderboards read on Nov 6, 2026. Gemini 4 Carbon and Claude Opus 5.5 score above GPT-6 Astra on Artificial Analysis's index, and Claude Opus 5.5 leads LMArena's WebDev board, though OpenAI's own tables show Astra ahead on many agent and math tests.
Where does OpenAI rank on LMArena? On the text board updated Oct 27, 2026, OpenAI's best model is GPT-5.6 Sol at 22nd, with GPT-6.1 Sol 27th and GPT-6 Astra 37th. On the WebDev board GPT-6 Astra is 2nd.
Which OpenAI model is best value? By Artificial Analysis's figures, GPT-6.1 Sol scores 55 for about $1.27 per index task, against 57 for $6.24 with GPT-6 Astra. GPT-6.1 Terra scores 51 for $0.96.
What is GPT-6 Astra's SWE-bench score? OpenAI has not published one, and the public SWE-bench leaderboards had no GPT-6 entry as of Nov 6, 2026.
Sources
- Artificial Analysis model leaderboard, read Nov 6, 2026
- Artificial Analysis, index v5 changelog, read Nov 6, 2026
- LMArena Text leaderboard, updated Oct 27, 2026, read Nov 6, 2026
- LMArena WebDev leaderboard, updated Oct 27, 2026, read Nov 6, 2026
- LMArena Vision leaderboard, updated Oct 19, 2026, read Nov 6, 2026
- LMArena leaderboard overview, read Nov 6, 2026
- SWE-bench Pro public leaderboard, Scale AI, read Nov 6, 2026
- SWE-bench leaderboard data, SWE-bench on GitHub, read Nov 6, 2026
- ARC Prize leaderboard, read Nov 6, 2026
- Introducing GPT-6 Astra, OpenAI, Sep 3, 2026
- GPT-6 Astra System Card, OpenAI Deployment Safety Hub, read Nov 6, 2026
- OpenAI unveils GPT-6.1 Sol at DevDay, Unite.AI, Sep 29, 2026
- OpenAI launches GPT-6.1 Terra, BleepingComputer, Oct 26, 2026
- Introducing Claude Opus 5.5, Anthropic, Sep 22, 2026
- Introducing Gemini 4 Argon, Google, Oct 13, 2026
- Codex vs Claude - an Honest Comparison, Tech With Tim, Aug 27, 2026 (video)
- GPT-6 Astra Is Finally Here (And It’s REALLY Good), Matt Wolfe, Sep 3, 2026 (video)
- Did OpenAI actually build AGI? GPT-6 Astra first look, Fireship, Sep 4, 2026 (video)
Moments
Codex vs Claude - an Honest Comparison
Tech With Tim
GPT-6 Astra Is Finally Here (And It’s REALLY Good)
Matt Wolfe
Did OpenAI actually build AGI? GPT-6 Astra first look
Fireship
Related
-
- GPT-6 LunaGPT-6 Luna is OpenAI's cheapest and fastest GPT-6 model, released Sep 22, 2026 at $0.10 per million input tokens and $0.50 output. OpenAI's pricing page and Oct 7 release note name it as ChatGPT's model for Free and Go, though its help articles say GPT-5.6 Luna.
-
-
- gpt-ossgpt-oss is OpenAI's family of open-weight models: gpt-oss-120b and gpt-oss-20b, released Aug 5, 2025 under the Apache 2.0 license, which run on one 80 GB GPU or a 16 GB machine. As of Nov 6, 2026 OpenAI has released no newer general open-weight model.
- GPT-Rosalind and Cyber ModelsOpenAI's specialized models are only for approved organizations: GPT-Rosalind for life sciences (GPT-6 Rosalind since Oct 30, 2026) and the Daybreak cyber models GPT-5.6 Cyber and GPT-6-Cyber (Nov 5, 2026), both GPT-6 versions built on GPT-6 Astra. They are in ChatGPT Work, Codex and the API, not the Chat tab.
-
OpenAI Reasoning ModelsOpenAI's reasoning models, the "o" series from o1-preview (Sep 12, 2024) through o3 and o4-mini (Apr 16, 2025), think before answering. GPT-5 merged them into its main models in August 2025, and today reasoning is a level you pick in ChatGPT (Medium to Extra High) or an effort setting in the API.
-
Retired ChatGPT ModelsEvery model OpenAI removed from ChatGPT, with dates: GPT-4o, GPT-4.1, o4-mini and GPT-5 on Feb 13, 2026, GPT-5.1 on Mar 11, GPT-5.2 on Jun 12, GPT-4.5 on Jun 26, o3 on Aug 26 and GPT-5.5 on Oct 14, 2026. Models generally stay 90 days after a successor arrives.
See also
-
-
-
ChatGPT vs ClaudeChatGPT and Claude cost the same $20 a month at their main paid tier, but they win at different tasks. Claude leads the public leaderboards for writing and coding as of Nov 6, 2026, while ChatGPT makes images, has the more generous free plan and offers cheaper and pricier plans that Claude lacks.
-
ChatGPT vs GeminiGoogle's Gemini ranks above ChatGPT on the public chat leaderboards as of Nov 6, 2026, costs about the same at $19.99 a month for Google AI Pro, makes video and comes with storage and Google app features. ChatGPT is ahead on coding agents and gives more on its free plan, and the two are close on most everyday tasks.
-
GPT-4oGPT-4o was OpenAI's "omni" model, launched May 13, 2024, that brought native voice and image generation to ChatGPT. It was retired from ChatGPT on Feb 13, 2026 (Apr 3, 2026 in business custom GPTs) and shuts down in the API on Jan 14, 2027.
-