Competition math, formal proofs, research-level problem solving, and verifiable reasoning. This report blends public leaderboard signals into one task-specific composite score, then highlights the practical cautions that matter before you choose a model.
Claude Fable 5
95.9
2026-06-12
Top contenders
Claude Fable 5
Model 98 · Confidence 90%
95.9
Gemini 3.5 Flash
Model 100 · Confidence 65%
91.2
Claude Opus 4.6 Thinking
Model 100 · Confidence 65%
91.2
GPT-5.4 High
Model 99 · Confidence 65%
90.5
Claude Opus 4.6
Model 99 · Confidence 65%
90.5
Full-source models
2
Average coverage
39%
Top-five spread
5.4
All ranked models
Showing all 21 models with at least one confirmed source row in this category. Models with no category source coverage are excluded. Confirmed rows are ordered by Bayesian-smoothed adjusted score; missing source rows stay n/a instead of counting as zero.
Best overall pick when formal proof and research math both matter.
Caution
Still verify every proof or derivation independently.
75% confirmed coverage · 90% confidence
Adjusted score
95.9
Model
98
Confidence
90%
Fast math assistance and high-volume quantitative problem solving.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
91.2
Model
100
Confidence
65%
Extended-reasoning math explanations and proof-style drafts.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
91.2
Model
100
Confidence
65%
High-effort OpenAI math reasoning and contest-style problem solving.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
90.5
Model
99
Confidence
65%
Strong math explanations and symbolic reasoning drafts.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
90.5
Model
99
Confidence
65%
Careful math reasoning where slower thinking-mode review is useful.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
90.5
Model
99
Confidence
65%
Long-context math prompts, multimodal math, and Google ecosystem workflows.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
90.5
Model
99
Confidence
65%
High-effort OpenAI math runs when exact reasoning settings can be controlled.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.9
Model
98
Confidence
65%
Premium math support and detailed solution review.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.9
Model
98
Confidence
65%
Thinking-mode solution planning and multi-step derivations.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.9
Model
98
Confidence
65%
Qwen math reasoning tests and cost-aware quantitative workflows.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.9
Model
98
Confidence
65%
Long-context math reading and open-weight oriented comparison.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.9
Model
98
Confidence
65%
Open-weight math shortlist testing and lower-cost deployment experiments.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.9
Model
98
Confidence
65%
Preview-model Qwen math comparisons.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.2
Model
97
Confidence
65%
Alternative math reasoning tests across structured prompts.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.2
Model
97
Confidence
65%
Fast math drafting and lower-latency reasoning workflows.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.2
Model
97
Confidence
65%
Reasoning-forward Kimi math prompts and long-context derivations.
Caution
Arena Math preference data is confirmed for this row; verify final answers with symbolic checks or your own benchmark set.
25% confirmed coverage · 65% confidence
Adjusted score
89.2
Model
97
Confidence
65%
Formal math and careful proof-style reasoning.
Caution
Less dominant than Fable 5 on Riemann-bench.
100% confirmed coverage · 100% confidence
Adjusted score
84.2
Model
84.2
Confidence
100%
Hard math plus general advanced reasoning.
Caution
Use exact model effort settings; benchmark names often include reasoning level.
75% confirmed coverage · 90% confidence
Adjusted score
82.5
Model
83.1
Confidence
90%
AIME-style math and broad reasoning workflows.
Caution
Formal proof benchmarks are a different skill from contest answer accuracy.
75% confirmed coverage · 90% confidence
Adjusted score
80.4
Model
80.8
Confidence
90%
Competition math, long context, and multimodal math workflows.
Caution
Riemann-bench and formal proof results trail the top Anthropic model in this snapshot.
100% confirmed coverage · 100% confidence
Adjusted score
78.3
Model
78.3
Confidence
100%
How to use it
The top model is the best blended pick for this query snapshot, but model choice should still account for price, latency, privacy, context length, tool access, safety settings, and your own benchmark prompts. Use this page to reduce the search space, then run a small evaluation on your real tasks before standardizing.
