Text ArenaLeaderboard
Answer the same prompt as a rival model; real users vote blind on which answer they prefer.
1,504
50% expected win rate against the board leader (1,507)
#3 of 92
max · board 2026-09-02
Source: ArenaAnthropicUnited States· released Sep 2026
Claude Fable 5.1
Best published result on 7 of the 17 tests it has taken.
Anthropic, United States. One of 12 models from this lab on the Index.
Third-party publishedMedium confidence (7+ comparable tests, 3+ Index categories)
Available in Theo as Theo Apex
Theo orchestrates it for the steps it does best and always shows which engine answered.
Snapshot September 8, 2026
OpenCharts Index
97.0
Ranked #2 of 36 · 3.0% behind the frontier on average · 17 comparable tests · 3 counting categories
Drop any single test and the Index lands between 96.6 and 97.8.
4 of 8 hardest tests · share of the frontier
Distance to today's frontier on the hardest tests. Not a measure of intelligence.
What it's known for
Claude Fable 5.1, in 17 results.
Strongest in reasoning (97.5) and coding (97.0). It holds the best published result on APEX-Agents, CursorBench and Humanity's Last Exam and 4 more. It trails the frontier most in math, 3.4% behind on average. No comparable results yet in multimodal and long context. Within 5% of the best published results, on average.
Best published result onAPEX-Agents47.4%CursorBench73.4%Humanity's Last Exam46.5%Mock AIME 2024–2025100.0%ProofBench100.0%SciCode62.0%WeirdML92.9%
Strongest category
Reasoning
Hard, novel problems: graduate science, abstract puzzles, expert exams.
97.5
- Humanity's Last Exam46.5% · best
- ARC-AGI-197.5% · 99% of best
- CritPt31.1% · 96% of best
- ARC-AGI-290.0% · 95% of best
Mean of 4 of the category's 6 comparable tests. Counts toward the Index.
Best single result
APEX-Agents
Complete professional-services work in consulting, law and finance as an agent, graded against expert rubrics.
47.4%
Best published result of 49 models.
Expert-graded knowledge work is what professional teams would actually delegate.
Where it trails
Vending-Bench 2
Run a simulated vending business for a year: ordering, pricing and cash flow. The score is the final balance.
$5,422
23.3% behind the best published result ($11,182) on a log scale · #20 of 53.
Long-horizon decisions whose consequences compound. It exposes models that drift or forget.
Not measured yet
No comparable result yet in these 2 categories. A blank is a blank, never a zero, and it does not lower the Index.
- Multimodal
- Long context
Covered, but not averaged into the Index:
- Knowledgesingle-test category, shown beside the Index · 93.7
- Agentic2 of 8 tests · needs 4 · mean 88.4
- Human preferencesingle-test category, shown beside the Index · 99.2
See it at work
What Claude Fable 5.1 was asked to do, and how it did.
Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.
Human preference · mean 99.2 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it
Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.
Head to head
Claude Fable 5.1 against whoever you pick.
Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.
Claude Fable 5.1
#2·Index 97.0A+
1 : 6
wins · 6 ties · 13 shared
GPT-6 Astra
#1·Index 99.0A+
GPT-6 Astra comes out ahead on 6 of the 13 tests both have taken.
Category means
Reasoning
97.599.6Knowledge
93.7100.0Coding
97.097.7Math
96.699.8Agentic
88.498.5Human preference
99.2—
Shared tests · biggest gaps first
- Code Arena (WebDev)Coding−10.01,7621,797
- −10.087.8%97.6%
- SciCodeCoding+9.062.0%56.5%
- SimpleQA VerifiedKnowledge−6.370.8%75.6%
- ARC-AGI-2Reasoning−5.390.0%95.0%
- FrontierCodeCoding−4.450.9%53.3%
- −3.790.2%93.7%
- CritPtReasoningtie31.1%31.7%
- APEX-AgentsAgentictie47.4%46.7%
- ARC-AGI-1Reasoningtie97.5%98.5%
- ProofBenchMathtie100.0%99.0%
- WeirdMLCodingtie92.9%92.9%
- tie100.0%100.0%
Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.
Open GPT-6 AstraScores
Every test, every source
The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.
Category means
- Reasoning97.5 · 4/6
- Knowledgebeside93.7 · 1/1
- Coding97.0 · 5/7
- Math96.6 · 4/5
- Agentic88.4 · 2/8
- Multimodal—
- Human preferencebeside99.2 · 1/1
- Long context—
Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.
Humanity's Last ExamLeaderboardReasoning
46.5%
best published result
of 18 models
xhigh
Source: Epoch AI — AI Benchmarking HubARC-AGI-2LeaderboardReasoning
90.0%
94.7% of the best (95.0%)
94th percentile of 51 · n = 120
max
Source: Epoch AI — AI Benchmarking HubARC-AGI-1LeaderboardReasoning
97.5%
99.0% of the best (98.5%)
90th percentile of 51 · n = 100
max
Source: Epoch AI — AI Benchmarking HubCritPtLeaderboardReasoning
31.1%
96.4% of the best (32.3%)
97th percentile of 78
xhigh
Source: Epoch AI — AI Benchmarking HubSimpleQA VerifiedEpoch-runKnowledge
70.8%
93.7% of the best (75.6%)
96th percentile of 57 · n = 1,000
max
Source: Epoch AI — AI Benchmarking HubSciCodeLeaderboardCoding
62.0%
best published result
of 73 models
max
Source: Epoch AI — AI Benchmarking HubWeirdMLLeaderboardCoding
92.9%
best published result
of 69 models
max
Source: Epoch AI — AI Benchmarking HubFrontierCodeLeaderboardCoding
50.9%
95.2% of the best (53.5%)
88th percentile of 27
medium · harness: claude-code · reasoning effort: medium
Source: Epoch AI — AI Benchmarking HubCursorBenchLeaderboardCoding
73.4%
best published result
of 21 models
max · reasoning level: Max
Source: Epoch AI — AI Benchmarking HubCode Arena (WebDev)LeaderboardCoding
1,762
45% expected win rate against the board leader (1,797)
99th percentile of 86
max · board 2026-09-05
Source: Arena (LMArena) — Leaderboard DatasetMock AIME 2024–2025Epoch-runMath
100.0%
best published result
of 78 models
max
Source: Epoch AI — AI Benchmarking HubFrontierMath (Tiers 1–3)Epoch-runMath
90.2%
96.3% of the best (93.7%)
98th percentile of 61 · n = 290
max
Source: Epoch AI — AI Benchmarking HubFrontierMath Tier 4Epoch-runMath
87.8%
90.0% of the best (97.6%)
96th percentile of 51 · n = 48
max
Source: Epoch AI — AI Benchmarking HubProofBenchLeaderboardMath
100.0%
best published result
of 61 models
max
Source: Epoch AI — AI Benchmarking HubAPEX-AgentsLeaderboardAgentic
47.4%
best published result
of 49 models
Vending-Bench 2LeaderboardAgentic
$5,422
76.7% of the best ($11,182) on a log scale
63th percentile of 53
Text ArenaLeaderboardHuman preference
1,504
50% expected win rate against the board leader (1,507)
98th percentile of 92
max · board 2026-09-02
Source: Arena (LMArena) — Leaderboard DatasetNearby on the leaderboard
Compare with its neighbours.
The best models, ranked here, working inside Theo.
28 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.