LMArena text
as of 28 Oct 2026Open trails by 48.7
DataAI benchmarks AI daily → explore →
AI benchmarks
95 benchmarks · 9,344 measured scores · 65 labs · 1 lab claims, not independently measured. Updated 1 h ago.
- Epoch AI Benchmarking Hubfetched 1 h ago · newest result 28 Oct 2026
- LMArenafetched 1 h ago · newest result 28 Oct 2026
- Lab-claimed (blogs and model cards)fetched 19 h ago · newest result 30 Oct 2026
- SWE-bench leaderboardsstalefetched 1 h ago · newest result 26 Feb 2026
not updating upstream (newest result 247 days old)
The race: Epoch Capabilities Index
Each line is one lab's best score so far, labelled at its end. The frontier is 169 (GPT-6 Astra, 3 Sep 2026).
Open weights against closed
The best open-weights model against the best closed one on the same benchmark. The time gap is how long ago the closed frontier first reached today's open-weights best.
GPQA diamond
as of 26 Oct 2026Open trails by 3.2 pts, about 8 months behind in time
OTIS Mock AIME 2024-2025
as of 26 Oct 2026Open trails by 0.6 pts, about 6 months behind in time
Epoch Capabilities Index
as of 26 Oct 2026Open trails by 10.2, about 8 months behind in time
ARC-AGI
as of 17 Oct 2026Open trails by 4.0 pts, about 5 months behind in time
ARC-AGI-2
as of 17 Oct 2026Open trails by 16.9 pts, about 8 months behind in time
Every benchmark: the best score over time
The line steps up each time a model beats every earlier one; each dot is coloured by the lab that set the record. Hollow diamonds are scores a lab claims that no leaderboard has measured.
LMArena text
as of 28 Oct 20261,540 Gemini 4 Carbon (High)
GPQA diamond
as of 26 Oct 202696.4% Gemini 4 Argon (high)
99% of the ceiling · +9.4 pts in 12 months · open weights ~8 months behind
OTIS Mock AIME 2024-2025
as of 26 Oct 2026100% GPT-5.5 Pro (xhigh)
100% of the ceiling · +8.6 pts in 12 months · open weights ~6 months behind
ARC-AGI
as of 17 Oct 202698.5% Claude Fable 5 (xhigh)
99% of the ceiling · +28.3 pts in 12 months · open weights ~5 months behind
ARC-AGI-2
as of 17 Oct 202695% GPT-6 Astra (max)
95% of the ceiling · +76.7 pts in 12 months · open weights ~8 months behind
CritPt
as of 28 Oct 202636% Gemini 4 Carbon (high)
36% of the ceiling · +23.4 pts in 12 months · open weights ~8 months behind
DTBench
as of 12 Oct 202698.9% Claude Opus 5.5 (max)
99% of the ceiling · +8.3 pts in 12 months · open weights ~6 months behind
SciCode
as of 28 Oct 202666.9% Claude Opus 5.5 (max)
67% of the ceiling · +20.6 pts in 12 months · open weights ~1 months behind
LMCA
as of 12 Oct 202668.3% Claude Sonnet 5.5 (max)
80% of the ceiling · +28.3 pts in 12 months · open weights ~6 months behind
SWE-bench Verified (agent submissions)
as of 26 Feb 202679.2% Claude 4.5 Opus
79% of the ceiling · +0.4 pts in 12 months · open weights ~7 months behind
Its source is not updating: read this as the last known state.
WeirdML
as of 3 Sep 202693.6% GPT-6 Astra (pro, max)
94% of the ceiling · +32.9 pts in 12 months · open weights ~3 months behind
LMArena vision
as of 19 Oct 20261,325 Claude Fable 5 (High)
Chess Puzzles
as of 17 Oct 202676% GPT-6 Astra (pro, max)
76% of the ceiling · +28.0 pts in 12 months · open weights ~11 months behind
LMArena web development
as of 28 Oct 20261,810 Claude Opus 5.5 (Max)
Mystery Game Puzzles
as of 26 Oct 202684% GPT-6 Astra (max)
84% of the ceiling · +55.0 pts in 12 months · open weights ~6 months behind
MMLU
as of 26 Dec 202488.1% GPT-4o (Nov 2024)
88% of the ceiling · 0.0 pts in 12 months · open weights ~2 months behind
No new results in 674 days.
WebDev Arena
as of 30 Sep 20261,820 Claude Opus 5.5 (max)
+401 in 12 months
ALE-Bench
as of 26 Oct 20262,951 GPT-6 Astra (max)
+1,789 in 12 months · open weights ~8 months behind
FrontierMath-Tiers-1-3-v2-Private
as of 26 Oct 202693.7% GPT-6 Astra (max)
99% of the ceiling · +37.9 pts in 12 months · open weights ~7 months behind
MATH level 5
as of 15 Oct 202598.1% GPT-5 (high)
98% of the ceiling · 0.0 pts in 12 months · open weights ~1 months behind
No new results in 381 days.
SimpleBench
as of 12 Aug 202681.9% Claude Fable 5 (max)
82% of the ceiling · +19.5 pts in 12 months · open weights ~14 months behind
FrontierMath-2025-02-28-Private
as of 28 May 202652.4% GPT-5.5 Pro (high)
92% of the ceiling · +20.0 pts in 12 months · open weights ~4 months behind
ForecastBench
as of 13 Aug 202667.0 GPT-5.5 (high)
+1.7 in 12 months · open weights ~17 months behind
SimpleQA Verified
as of 26 Oct 202676% Gemini 4 Argon (high)
76% of the ceiling · +25.9 pts in 12 months · open weights ~10 months behind
67 more benchmarks
GSM8K
as of 29 Nov 202494.5% DeepSeek-Coder-V2 236B
95% of the ceiling · 0.0 pts in 12 months
No new results in 701 days.
SWE-bench Lite (agent submissions)
as of 11 Sep 202560.3% Claude 4 Sonnet
60% of the ceiling · 0.0 pts in 12 months · open weights ~8 months behind
Its source is not updating: read this as the last known state.
CursorBench
as of 17 Oct 202657.8% Claude Opus 5.5 (max)
58% of the ceiling · open weights ~2 months behind
ProofBench
as of 23 Oct 2026100% Claude Fable 5.1 (max)
100% of the ceiling · +81.0 pts in 12 months · open weights ~1 months behind
Winogrande
as of 26 Dec 202489.2% Llama 3.1-405B
89% of the ceiling · 0.0 pts in 12 months
No new results in 674 days.
FrontierMath-Tier-4-v2-Private
as of 26 Oct 2026100% GPT-6.1 Sol (max)
100% of the ceiling · +78.0 pts in 12 months · open weights ~8 months behind
ARC AI2
as of 26 Dec 202495.3% Llama 3.1-405B
95% of the ceiling · 0.0 pts in 12 months
No new results in 674 days.
FrontierMath-Tier-4-2025-07-01-Private
as of 28 May 202647.9% AI co-mathematician
80% of the ceiling · +33.3 pts in 12 months · open weights ~7 months behind
Aider polyglot
as of 1 Dec 202588% GPT-5 (high)
88% of the ceiling · 0.0 pts in 12 months · open weights ~6 months behind
No new results in 334 days.
Vending-Bench 2
as of 28 Oct 202616,484 Gemini 4 Carbon (unknown)
+12,645 in 12 months · open weights ~2 months behind
HellaSwag
as of 26 Dec 202495.3% GPT-4 (Mar 2023)
95% of the ceiling · 0.0 pts in 12 months · open weights ~17 months behind
No new results in 674 days.
DeepSWE
as of 3 Sep 202674.1% GPT-6 Astra (xhigh)
74% of the ceiling · open weights ~2 months behind
BoolQ
as of 17 Aug 202490.9% PaLM 2-L
91% of the ceiling · 0.0 pts in 12 months
No new results in 805 days.
Fiction.LiveBench
as of 27 Jan 202697.2% o3-pro
97% of the ceiling · 0.0 pts in 12 months · open weights ~10 months behind
No new results in 277 days.
APEX-Agents
as of 26 Oct 202682.4% Gemini 4 Argon (unknown)
82% of the ceiling · +78.0 pts in 12 months
PIQA
as of 26 Dec 202488.7% GPT-4o mini
89% of the ceiling · 0.0 pts in 12 months
No new results in 674 days.
Terminal Bench
as of 23 Apr 202681.8% GPT-5.4 (unknown thinking)
82% of the ceiling · +40.5 pts in 12 months · open weights ~3 months behind
No new results in 191 days.
GDP.pdf
as of 12 Oct 202635% GPT-6.1 Sol (max)
35% of the ceiling · open weights ~5 months behind
HLE
as of 7 Oct 202654.8% GPT-6 Astra (unknown thinking)
55% of the ceiling · +23.2 pts in 12 months · open weights ~6 months behind
BBH
as of 26 Dec 202489.2% Gemini 1.5 Pro (May 2024)
89% of the ceiling · 0.0 pts in 12 months · open weights ~8 months behind
No new results in 674 days.
LiveBench
as of 13 Nov 202582.3% Gemini 2.5 Pro Exp (Mar 2025)
82% of the ceiling · 0.0 pts in 12 months · open weights ~3 months behind
No new results in 352 days.
VideoMME
as of 18 Jun 202579.7% video-SALMONN 2+
80% of the ceiling · 0.0 pts in 12 months · open weights ~7 months behind
No new results in 500 days.
Lech Mazur Writing
as of 7 Aug 202586% GPT-5 (medium)
86% of the ceiling · 0.0 pts in 12 months
No new results in 450 days.
EnigmaEval
as of 9 Jul 202639.3% Claude Fable 5 (high)
39% of the ceiling · +20.5 pts in 12 months · open weights ~14 months behind
FrontierCode
as of 7 Oct 202654.6% Claude Opus 5.5 (medium)
55% of the ceiling · open weights ~2 months behind
METR Time Horizons
as of 7 Apr 202617.4 h Claude Mythos Preview (Early)
+842 in 12 months · open weights ~9 months behind
No new results in 207 days.
GSO-Bench
as of 3 Sep 202688.2% Claude Fable 5.1 (unknown)
88% of the ceiling · +73.5 pts in 12 months · open weights ~3 months behind
LMArena search
as of 13 Oct 20261,261 gpt-5.6-sol-xhigh
OpenBookQA
as of 23 Apr 202488% phi-3-small 7.4B
88% of the ceiling · 0.0 pts in 12 months
No new results in 921 days.
DeepResearch Bench
as of 28 May 202655.3% Claude Opus 4.6 (high)
55% of the ceiling · +2.7 pts in 12 months · open weights ~3 months behind
Balrog
as of 3 Sep 202668.3% GPT-6 Astra (max)
68% of the ceiling · +24.7 pts in 12 months
TriviaQA
as of 26 Dec 202487.5% Claude 2
88% of the ceiling · 0.0 pts in 12 months · open weights ~21 months behind
No new results in 674 days.
VPCT
as of 17 Dec 202591% Gemini 3 Pro Preview
91% of the ceiling · +25.0 pts in 12 months
No new results in 318 days.
Blueprint-Bench 2
as of 28 Oct 202658.6% Gemini 4 Carbon (unknown)
59% of the ceiling · +58.6 pts in 12 months · open weights ~3 months behind
Furniture Assembly
as of 26 Oct 202683.3% Claude Opus 5.5 (max)
83% of the ceiling · open weights ~7 months behind
SWE-Bench verified
as of 16 Jun 202683.5% Claude Opus 4.7 (max)
83% of the ceiling · +9.9 pts in 12 months · open weights ~2 months behind
GeoBench
as of 17 Dec 202588% Gemini 3 Flash Preview
88% of the ceiling · +7.0 pts in 12 months · open weights ~2 months behind
No new results in 318 days.
MirrorCode
as of 12 Oct 202697.8% Claude Sonnet 5.5 (max)
98% of the ceiling · +93.3 pts in 12 months · open weights ~2 months behind
Surface Evolver Bench
as of 6 Oct 202695% Claude Fable 5 (high)
95% of the ceiling · +70.0 pts in 12 months · open weights ~1 months behind
EBR-bench
as of 26 Oct 202678.1% GPT-6.1 Terra (max)
78% of the ceiling · +65.4 pts in 12 months · open weights ~10 months behind
LAMBADA
as of 30 Nov 202379.8% Falcon-180B
80% of the ceiling · 0.0 pts in 12 months
No new results in 1066 days.
WeirdML v3
as of 26 Oct 202654% GPT-6 Astra (xhigh)
60% of the ceiling · open weights ~4 months behind
SWE-bench Full (agent submissions)
as of 19 Dec 202552.6% Claude 4.5 Opus
53% of the ceiling · +8.4 pts in 12 months
Its source is not updating: read this as the last known state.
GBAEval
as of 24 Jul 202679.6% Claude Opus 5 (unknown)
80% of the ceiling · open weights ~5 months behind
ScienceQA
as of 16 Aug 202491.3% Phi-3.5-vision-instruct
91% of the ceiling · 0.0 pts in 12 months
No new results in 806 days.
CL-bench
as of 31 Mar 202627.9% GPT-5.4 (xhigh)
28% of the ceiling · +10.1 pts in 12 months · open weights ~3 months behind
No new results in 214 days.
Cybench
as of 5 Feb 202693% Claude Opus 4.6 (unknown thinking)
93% of the ceiling · +33.0 pts in 12 months · open weights ~2 months behind
No new results in 268 days.
SWE-bench Multimodal (agent submissions)
as of 17 Nov 202536% o3
36% of the ceiling · 0.0 pts in 12 months
Its source is not updating: read this as the last known state.
FrontierSWE
as of 7 Oct 202665.5% GPT-6 Astra (max)
66% of the ceiling · open weights ~2 months behind
AlgoTune
as of 5 Mar 20262.0 GPT-5.2 (medium)
+0.3 in 12 months
No new results in 240 days.
Remote Labor Index
as of 30 Sep 202621.3% Claude Opus 5.5 (unknown)
21% of the ceiling · +19.2 pts in 12 months
CL-bench Life
as of 24 Apr 202622.2% GPT-5.5 (high)
22% of the ceiling · +12.7 pts in 12 months · open weights ~5 months behind
No new results in 190 days.
CadEval
as of 16 Apr 202574% o3 (medium)
74% of the ceiling · 0.0 pts in 12 months
No new results in 563 days.
FrontierMath-Erdos
as of 30 Sep 20264.4% GPT-6.1 Sol (max)
4% of the ceiling · open weights ~3 months behind
OSWorld 2.0
as of 24 Jul 202631.4% Claude Opus 5 (max)
31% of the ceiling · open weights ~2 months behind
The Agent Company
as of 29 Sep 202542.9% DeepSeek-V3.2-Exp
43% of the ceiling · 0.0 pts in 12 months
No new results in 397 days.
SWE-bench Multilingual (agent submissions)
as of 20 Feb 202672.7% Gemini 3 Flash
73% of the ceiling
Its source is not updating: read this as the last known state.
PostTrainBench
as of 24 Jul 202641.8% Claude Fable 5 (max)
42% of the ceiling · open weights ~2 months behind
ANLI
as of 23 Apr 202458.1% GPT-3.5 Turbo (Nov 2023)
58% of the ceiling · 0.0 pts in 12 months · open weights ~6 months behind
No new results in 921 days.
GDPval
as of 11 Dec 202549.7% GPT-5.2 (none)
50% of the ceiling · +6.1 pts in 12 months
No new results in 324 days.
ExploitBench
as of 23 Apr 202670.6% Claude Mythos Preview (Early)
71% of the ceiling · +57.2 pts in 12 months · open weights ~2 months behind
No new results in 191 days.
OSWorld
as of 17 Feb 202672.1% Claude Sonnet 4.6 (no thinking)
72% of the ceiling · +29.2 pts in 12 months · open weights ~2 months behind
No new results in 256 days.
SpatialViz-Bench
as of 17 Jun 202544.7% Gemini 2.5 Pro (Jun 2025)
45% of the ceiling · 0.0 pts in 12 months · open weights ~4 months behind
No new results in 501 days.
SpireClimb
as of 29 Sep 202625.7% GPT-6 Astra (max)
26% of the ceiling
SpireClimb (Slay the Spire 1, Ascension 20 Heart)
as of 29 Sep 202620% GPT-6.1 Sol (max)
20% of the ceiling
SpireClimb (Slay the Spire 2, Ascension 10)
as of 29 Sep 202634.7% GPT-6 Astra (max)
35% of the ceiling
MindCube
as of 22 May 202546.7% Gemma 3 12B
47% of the ceiling · 0.0 pts in 12 months
No new results in 527 days.
Latest scores
One row per benchmark: its current leader. Click a heading to sort. "Of ceiling" is the leader's score as a share of the best achievable score.
| Leader | Lab | Weights | |||||
|---|---|---|---|---|---|---|---|
| LMArena text1,540 | Gemini 4 Carbon (High) | closed | 809 | n/a | n/a | 28 Oct 2026 | |
| GPQA diamond96.4% | Gemini 4 Argon (high) | closed | 326 | 99% | +9.4 pts | 26 Oct 2026 | |
| OTIS Mock AIME 2024-2025100% | GPT-5.5 Pro (xhigh) | OpenAI | closed | 306 | 100% | +8.6 pts | 26 Oct 2026 |
| Epoch Capabilities Index169 | GPT-6 Astra | OpenAI | closed | 282 | n/a | +18.3 | 26 Oct 2026 |
| ARC-AGI98.5% | Claude Fable 5 (xhigh) | Anthropic | closed | 278 | 99% | +28.3 pts | 17 Oct 2026 |
| ARC-AGI-295% | GPT-6 Astra (max) | OpenAI | closed | 266 | 95% | +76.7 pts | 17 Oct 2026 |
| CritPt36% | Gemini 4 Carbon (high) | closed | 256 | 36% | +23.4 pts | 28 Oct 2026 | |
| DTBench98.9% | Claude Opus 5.5 (max) | Anthropic | closed | 233 | 99% | +8.3 pts | 12 Oct 2026 |
| SciCode66.9% | Claude Opus 5.5 (max) | Anthropic | closed | 226 | 67% | +20.6 pts | 28 Oct 2026 |
| LMCA68.3% | Claude Sonnet 5.5 (max) | Anthropic | closed | 195 | 80% | +28.3 pts | 12 Oct 2026 |
| SWE-bench Verified (agent submissions)79.2% | Claude 4.5 Opus | Anthropic | closed | 180 | 79% | +0.4 pts | 26 Feb 2026 |
| WeirdML93.6% | GPT-6 Astra (pro, max) | OpenAI | closed | 170 | 94% | +32.9 pts | 3 Sep 2026 |
| LMArena vision1,325 | Claude Fable 5 (High) | Anthropic | closed | 167 | n/a | n/a | 19 Oct 2026 |
| Chess Puzzles76% | GPT-6 Astra (pro, max) | OpenAI | closed | 154 | 76% | +28.0 pts | 17 Oct 2026 |
| LMArena web development1,810 | Claude Opus 5.5 (Max) | Anthropic | closed | 153 | n/a | n/a | 28 Oct 2026 |
| Mystery Game Puzzles84% | GPT-6 Astra (max) | OpenAI | closed | 142 | 84% | +55.0 pts | 26 Oct 2026 |
| MMLU88.1% | GPT-4o (Nov 2024) | OpenAI | closed | 135 | 88% | 0.0 pts | 26 Dec 2024 |
| WebDev Arena1,820 | Claude Opus 5.5 (max) | Anthropic | closed | 129 | n/a | +401 | 30 Sep 2026 |
| ALE-Bench2,951 | GPT-6 Astra (max) | OpenAI | closed | 127 | n/a | +1,789 | 26 Oct 2026 |
| FrontierMath-Tiers-1-3-v2-Private93.7% | GPT-6 Astra (max) | OpenAI | closed | 122 | 99% | +37.9 pts | 26 Oct 2026 |
| MATH level 598.1% | GPT-5 (high) | OpenAI | closed | 108 | 98% | 0.0 pts | 15 Oct 2025 |
| SimpleBench81.9% | Claude Fable 5 (max) | Anthropic | closed | 102 | 82% | +19.5 pts | 12 Aug 2026 |
| FrontierMath-2025-02-28-Private52.4% | GPT-5.5 Pro (high) | OpenAI | closed | 101 | 92% | +20.0 pts | 28 May 2026 |
| ForecastBench67.0 | GPT-5.5 (high) | OpenAI | closed | 97 | n/a | +1.7 | 13 Aug 2026 |
| SimpleQA Verified76% | Gemini 4 Argon (high) | closed | 93 | 76% | +25.9 pts | 26 Oct 2026 | |
| GSM8K94.5% | DeepSeek-Coder-V2 236B | DeepSeek | open | 85 | 95% | 0.0 pts | 29 Nov 2024 |
| SWE-bench Lite (agent submissions)60.3% | Claude 4 Sonnet | Anthropic | closed | 84 | 60% | 0.0 pts | 11 Sep 2025 |
| ProofBench100% | Claude Fable 5.1 (max) | Anthropic | closed | 82 | 100% | +81.0 pts | 23 Oct 2026 |
| CursorBench57.8% | Claude Opus 5.5 (max) | Anthropic | closed | 82 | 58% | n/a | 17 Oct 2026 |
| Winogrande89.2% | Llama 3.1-405B | Meta | open | 80 | 89% | 0.0 pts | 26 Dec 2024 |
| FrontierMath-Tier-4-v2-Private100% | GPT-6.1 Sol (max) | OpenAI | closed | 78 | 100% | +78.0 pts | 26 Oct 2026 |
| ARC AI295.3% | Llama 3.1-405B | Meta | open | 76 | 95% | 0.0 pts | 26 Dec 2024 |
| FrontierMath-Tier-4-2025-07-01-Private47.9% | AI co-mathematician | closed | 72 | 80% | +33.3 pts | 28 May 2026 | |
| Vending-Bench 216,484 | Gemini 4 Carbon (unknown) | closed | 71 | n/a | +12,645 | 28 Oct 2026 | |
| Aider polyglot88% | GPT-5 (high) | OpenAI | closed | 71 | 88% | 0.0 pts | 1 Dec 2025 |
| HellaSwag95.3% | GPT-4 (Mar 2023) | OpenAI | closed | 70 | 95% | 0.0 pts | 26 Dec 2024 |
| DeepSWE74.1% | GPT-6 Astra (xhigh) | OpenAI | closed | 69 | 74% | n/a | 3 Sep 2026 |
| BoolQ90.9% | PaLM 2-L | Unknown | n/a | 65 | 91% | 0.0 pts | 17 Aug 2024 |
| Fiction.LiveBench97.2% | o3-pro | OpenAI | closed | 61 | 97% | 0.0 pts | 27 Jan 2026 |
| APEX-Agents82.4% | Gemini 4 Argon (unknown) | closed | 60 | 82% | +78.0 pts | 26 Oct 2026 | |
| Terminal Bench81.8% | GPT-5.4 (unknown thinking) | OpenAI | closed | 59 | 82% | +40.5 pts | 23 Apr 2026 |
| PIQA88.7% | GPT-4o mini | OpenAI | closed | 59 | 89% | 0.0 pts | 26 Dec 2024 |
| GDP.pdf35% | GPT-6.1 Sol (max) | OpenAI | closed | 56 | 35% | n/a | 12 Oct 2026 |
| HLE54.8% | GPT-6 Astra (unknown thinking) | OpenAI | closed | 51 | 55% | +23.2 pts | 7 Oct 2026 |
| VideoMME79.7% | video-SALMONN 2+ | ByteDance | n/a | 50 | 80% | 0.0 pts | 18 Jun 2025 |
| LiveBench82.3% | Gemini 2.5 Pro Exp (Mar 2025) | closed | 50 | 82% | 0.0 pts | 13 Nov 2025 | |
| BBH89.2% | Gemini 1.5 Pro (May 2024) | closed | 50 | 89% | 0.0 pts | 26 Dec 2024 | |
| Lech Mazur Writing86% | GPT-5 (medium) | OpenAI | closed | 49 | 86% | 0.0 pts | 7 Aug 2025 |
| FrontierCode54.6% | Claude Opus 5.5 (medium) | Anthropic | closed | 46 | 55% | n/a | 7 Oct 2026 |
| EnigmaEval39.3% | Claude Fable 5 (high) | Anthropic | closed | 46 | 39% | +20.5 pts | 9 Jul 2026 |
| METR Time Horizons17.4 h | Claude Mythos Preview (Early) | Anthropic | closed | 45 | n/a | +842 | 7 Apr 2026 |
| OpenBookQA88% | phi-3-small 7.4B | Microsoft | open | 42 | 88% | 0.0 pts | 23 Apr 2024 |
| LMArena search1,261 | gpt-5.6-sol-xhigh | OpenAI | closed | 42 | n/a | n/a | 13 Oct 2026 |
| GSO-Bench88.2% | Claude Fable 5.1 (unknown) | Anthropic | closed | 42 | 88% | +73.5 pts | 3 Sep 2026 |
| DeepResearch Bench55.3% | Claude Opus 4.6 (high) | Anthropic | closed | 41 | 55% | +2.7 pts | 28 May 2026 |
| Balrog68.3% | GPT-6 Astra (max) | OpenAI | closed | 40 | 68% | +24.7 pts | 3 Sep 2026 |
| VPCT91% | Gemini 3 Pro Preview | closed | 38 | 91% | +25.0 pts | 17 Dec 2025 | |
| TriviaQA87.5% | Claude 2 | Anthropic | closed | 38 | 88% | 0.0 pts | 26 Dec 2024 |
| Furniture Assembly83.3% | Claude Opus 5.5 (max) | Anthropic | closed | 35 | 83% | n/a | 26 Oct 2026 |
| Blueprint-Bench 258.6% | Gemini 4 Carbon (unknown) | closed | 35 | 59% | +58.6 pts | 28 Oct 2026 | |
| SWE-Bench verified83.5% | Claude Opus 4.7 (max) | Anthropic | closed | 33 | 83% | +9.9 pts | 16 Jun 2026 |
| MirrorCode97.8% | Claude Sonnet 5.5 (max) | Anthropic | closed | 31 | 98% | +93.3 pts | 12 Oct 2026 |
| GeoBench88% | Gemini 3 Flash Preview | closed | 31 | 88% | +7.0 pts | 17 Dec 2025 | |
| Surface Evolver Bench95% | Claude Fable 5 (high) | Anthropic | closed | 29 | 95% | +70.0 pts | 6 Oct 2026 |
| EBR-bench78.1% | GPT-6.1 Terra (max) | OpenAI | closed | 28 | 78% | +65.4 pts | 26 Oct 2026 |
| LAMBADA79.8% | Falcon-180B | Technology Innovation Institute | open | 26 | 80% | 0.0 pts | 30 Nov 2023 |
| WeirdML v354% | GPT-6 Astra (xhigh) | OpenAI | closed | 25 | 60% | n/a | 26 Oct 2026 |
| SWE-bench Full (agent submissions)52.6% | Claude 4.5 Opus | Anthropic | closed | 24 | 53% | +8.4 pts | 19 Dec 2025 |
| ScienceQA91.3% | Phi-3.5-vision-instruct | Unknown | n/a | 23 | 91% | 0.0 pts | 16 Aug 2024 |
| GBAEval79.6% | Claude Opus 5 (unknown) | Anthropic | closed | 23 | 80% | n/a | 24 Jul 2026 |
| SWE-bench Multimodal (agent submissions)36% | o3 | OpenAI | closed | 22 | 36% | 0.0 pts | 17 Nov 2025 |
| Cybench93% | Claude Opus 4.6 (unknown thinking) | Anthropic | closed | 22 | 93% | +33.0 pts | 5 Feb 2026 |
| CL-bench27.9% | GPT-5.4 (xhigh) | OpenAI | closed | 22 | 28% | +10.1 pts | 31 Mar 2026 |
| FrontierSWE65.5% | GPT-6 Astra (max) | OpenAI | closed | 21 | 66% | n/a | 7 Oct 2026 |
| Remote Labor Index21.3% | Claude Opus 5.5 (unknown) | Anthropic | closed | 18 | 21% | +19.2 pts | 30 Sep 2026 |
| AlgoTune2.0 | GPT-5.2 (medium) | OpenAI | closed | 18 | n/a | +0.3 | 5 Mar 2026 |
| CL-bench Life22.2% | GPT-5.5 (high) | OpenAI | closed | 16 | 22% | +12.7 pts | 24 Apr 2026 |
| The Agent Company42.9% | DeepSeek-V3.2-Exp | DeepSeek | open | 14 | 43% | 0.0 pts | 29 Sep 2025 |
| OSWorld 2.031.4% | Claude Opus 5 (max) | Anthropic | closed | 14 | 31% | n/a | 24 Jul 2026 |
| FrontierMath-Erdos4.4% | GPT-6.1 Sol (max) | OpenAI | closed | 14 | 4% | n/a | 30 Sep 2026 |
| CadEval74% | o3 (medium) | OpenAI | closed | 14 | 74% | 0.0 pts | 16 Apr 2025 |
| SWE-bench Multilingual (agent submissions)72.7% | Gemini 3 Flash | closed | 13 | 73% | n/a | 20 Feb 2026 | |
| PostTrainBench41.8% | Claude Fable 5 (max) | Anthropic | closed | 12 | 42% | n/a | 24 Jul 2026 |
| GDPval49.7% | GPT-5.2 (none) | OpenAI | closed | 11 | 50% | +6.1 pts | 11 Dec 2025 |
| ANLI58.1% | GPT-3.5 Turbo (Nov 2023) | OpenAI | closed | 11 | 58% | 0.0 pts | 23 Apr 2024 |
| OSWorld72.1% | Claude Sonnet 4.6 (no thinking) | Anthropic | closed | 9 | 72% | +29.2 pts | 17 Feb 2026 |
| ExploitBench70.6% | Claude Mythos Preview (Early) | Anthropic | closed | 9 | 71% | +57.2 pts | 23 Apr 2026 |
| SpatialViz-Bench44.7% | Gemini 2.5 Pro (Jun 2025) | closed | 8 | 45% | 0.0 pts | 17 Jun 2025 | |
| SpireClimb (Slay the Spire 2, Ascension 10)34.7% | GPT-6 Astra (max) | OpenAI | closed | 7 | 35% | n/a | 29 Sep 2026 |
| SpireClimb (Slay the Spire 1, Ascension 20 Heart)20% | GPT-6.1 Sol (max) | OpenAI | closed | 7 | 20% | n/a | 29 Sep 2026 |
| SpireClimb25.7% | GPT-6 Astra (max) | OpenAI | closed | 7 | 26% | n/a | 29 Sep 2026 |
| MindCube46.7% | Gemma 3 12B | open | 5 | 47% | 0.0 pts | 22 May 2025 | |
| CSQA257% | GPT-3.5 Turbo (June 2023) | OpenAI | closed | 3 | 57% | 0.0 pts | 18 Jul 2023 |
| SuperGLUE71.8% | InstructGPT 175B | OpenAI | closed | 1 | 72% | 0.0 pts | 27 Jan 2022 |
Claimed by the labs, not independently measured
Scores found in a lab's own announcement or model card, extracted automatically once a day. They appear days before any leaderboard runs the model and are the lab's word, so they are kept out of every line above.
- Gemini 4 Carbon 88.2% on SEC-bench Pro, Google (30 Oct 2026) source
Sources: Epoch AI Benchmarking Hub (CC BY 4.0, credit Epoch AI); LMArena (CC BY 4.0 dataset (lmarena-ai/leaderboard-dataset on Hugging Face)); SWE-bench leaderboards (public GitHub data; credit SWE-bench). A score is placed by the model's release date (or the leaderboard's publish date for Elo ratings, which we have captured since our first fetch). Machine access: /api/v1/ai-benchmarks.