CtrlAltDebrief

AI model benchmarks

25 benchmarks across reasoning, maths, coding and agentic tasks. Each one ranks every model that carries a score and puts its API cost alongside, where it has one.

Benchmarks

25

leaderboards

Categories

5

reasoning · maths · coding · agents

Models ranked

485

on the widest board

Source

AA + HF

benchmark scores

Headline indices

3 benchmarks
Intelligence IndexComposite
485 models ranked
  1. 1Claude Opus 5.557.6
  2. 2Claude Sonnet 5.556.0
  3. 3Muse Wonder Full (Max)55.8
View full leaderboard →
Coding IndexCompositeFrozen Sep 9, 2026
185 models ranked
  1. 1Claude Fable 5.181.6
  2. 2GPT-5.6 Sol78.3
  3. 3Claude Opus 578.0
View full leaderboard →
Agentic IndexCompositeFrozen Sep 9, 2026
160 models ranked
  1. 1Claude Fable 5.158.0
  2. 2Qwen3.8-Flash-Next56.4
  3. 3Claude Opus 556.2
View full leaderboard →

Reasoning & knowledge

7 benchmarks
Humanity's Last Exam
460 models ranked
  1. 1Claude Opus 5.561%
  2. 2Claude Fable 5.159%
  3. 3Gemini 4 Argon57%
View full leaderboard →
GPQA Diamond
448 models ranked
  1. 1GPT-6 Astra96%
  2. 2Gemini 3.8 Flash95%
  3. 3Grok 4.695%
View full leaderboard →
CritPt
380 models ranked
  1. 1Muse Wonder Full (Max)33%
  2. 2GPT-5.6 Sol32%
  3. 3Claude Opus 5.532%
View full leaderboard →
Omniscience
377 models ranked
  1. 1Claude Opus 5.546.4
  2. 2Claude Fable 5.143.5
  3. 3GPT-6 Astra43.4
View full leaderboard →
MMMU-Pro
164 models ranked
  1. 1Claude Opus 5.588%
  2. 2GPT-6 Astra87%
  3. 3GPT-6.1 Sol86%
View full leaderboard →
ARC-AGI-2
63 models ranked
  1. 1GPT-5.6 Sol93%
  2. 2GPT-5.585%
  3. 3GPT-5.5 Pro85%
View full leaderboard →
MMLU-ProFrozen Aug 18, 2026
267 models ranked
  1. 1Gemini 3 Pro90%
  2. 2Claude Opus 4.589%
  3. 3Gemini 3 Flash89%
View full leaderboard →

Math

1 benchmarks
AIME 2026
20 models ranked
  1. 1Inkling97%
  2. 2A.X-K297%
  3. 3Step 3.5 Flash97%
View full leaderboard →

Coding & agents

12 benchmarks
τ²-bench
315 models ranked
  1. 1GLM 5.299%
  2. 2JT-35B-Flash99%
  3. 3GLM 4.7 Flash99%
View full leaderboard →
TerminalBench Hard
309 models ranked
  1. 1GPT-5.6 Sol66%
  2. 2Claude Fable 563%
  3. 3GPT-5.6 Terra63%
View full leaderboard →
TerminalBench v2.1
174 models ranked
  1. 1Claude Fable 5.191%
  2. 2GPT-5.6 Sol90%
  3. 3Claude Opus 589%
View full leaderboard →
τ-bench Banking
147 models ranked
  1. 1Qwen3.8 Max51%
  2. 2Grok 4.651%
  3. 3Muse Spark 1.351%
View full leaderboard →
LiveCodeBench
144 models ranked
  1. 1Gemini 3 Pro92%
  2. 2Gemini 3 Flash91%
  3. 3GLM 4.789%
View full leaderboard →
SciCode
144 models ranked
  1. 1Claude Opus 5.567%
  2. 2Claude Fable 5.163%
  3. 3Muse Wonder Full (Max)63%
View full leaderboard →
IT-Bench SRE
43 models ranked
  1. 1GPT-5.6 Sol56%
  2. 2Step 5 Preview56%
  3. 3Gemini 3.8 Flash53%
View full leaderboard →
SWE-bench Verified
37 models ranked
  1. 1Macaron V1 Venti86%
  2. 2DeepSeek V4 Pro 042381%
  3. 3MiniMax M381%
View full leaderboard →
SWE-bench Pro
34 models ranked
  1. 1Qwen3.8 2.4T A95B68%
  2. 2Hy4 preview66%
  3. 3Qwen3.8-Flash-Next63%
View full leaderboard →
APEX Agents
33 models ranked
  1. 1Gemini 3.5 Flash47%
  2. 2Qwen3.8 Max (0902)42%
  3. 3Kimi K341%
View full leaderboard →
AA-BriefcaseUndated
24 models ranked
  1. 1Claude Fable 554%
  2. 2Claude Sonnet 545%
  3. 3Claude Opus 4.843%
View full leaderboard →
GDPvalFrozen Aug 18, 2026
202 models ranked
  1. 1Claude Opus 51844.7
  2. 2Grok 4.61747.5
  3. 3Claude Fable 51737.7
View full leaderboard →

Instruction & long-context

2 benchmarks
Long-Context ReasoningComposite
379 models ranked
  1. 1Kimi K389%
  2. 2Step 5 Preview88%
  3. 3MiMo-V2.6-Pro86%
View full leaderboard →
IFBench
322 models ranked
  1. 1Grok 4.383%
  2. 2Grok 4 20 030983%
  3. 3MiniMax M383%
View full leaderboard →
Benchmark scores via ARC Prize, Artificial Analysis and Hugging Face leaderboards · Listed prices from LiteLLM, OpenRouter and Vercel

Every weekday

AI moves fast. Here's your debrief.

News, analysis, tools, and more.

For people who build with AI