Live · updated 15 Oct 2026
One scorefrom every benchmark that matters.
Lab tests, blind human votes, fresh tasks no model has seen, and the tests creators run themselves, boiled down to a single number. Real use counts double, so a model can't top B.O.B. just by acing the usual exams.
- 95
- models ranked
- 46
- benchmarks
- 11
- real-use tests
- 12
- sources
Fed by Epoch AI · LMArena · LiveBench · Vals AI · Artificial Analysis · BridgeBench · SWE-rebench · Kagi · EQ-Bench · creator tests. How it's scored
Rank 1 · B.O.B.
Claude Opus 5.5
Anthropic · 31 tests
95.9
- Real use97
- Reasoning90
- Math96
- Coding98
- Agents94
- Indexes100
- #1Claude Opus 5.5 · 95.9/
- #2GPT-6 Astra · 94.2/
- #3Claude Fable 5.1 · 91.6/
- Reality checkGPT-5.6 Terra: lab 71 vs real use 42/
- Reality checkStep 5 Preview: lab 70 vs real use 43/
- Reality checkGPT-5.6 Luna: lab 53 vs real use 31/
- Reality checkDeepSeek V4 Pro: lab 52 vs real use 31/
- PriceGemma 4 26B A4B now $0.076 in / $0.26 out/
- PriceMiniMax-M2.5 now $0.27 in / $0.95 out/
- PriceQwen3.6 35B A3B now $0.15 in / $1.00 out/
- Leak · rumorClaude Fable 5.5/
- Leak · spottedA stronger model behind Claude Opus 5.5 requests/
- Leak · rumor"Claude Next" inside Claude Code/
- NewsOpenAI Launches Amp to Automate Complex Workflows Beyond Code/
- NewsLiquid AI Drops LFM2.5-24B Non-Transformer MoE Model/
- NewsCloudflare Releases Clef Open Multimodal Model for the Edge/
01Podium
The top three right now
Claude Opus 5.5
Anthropic · 31 tests
Real use
Reasoning
Math
Coding
Agents
Indexes
$4.00 in · $20.00 out per 1M tokens
GPT-6 Astra
OpenAI · 34 tests
Real use
Reasoning
Math
Coding
Agents
Indexes
$10.00 in · $50.00 out per 1M tokens
Claude Fable 5.1
Anthropic · 33 tests
Real use
Reasoning
Math
Coding
Agents
Indexes
$10.00 in · $50.00 out per 1M tokens
02Leaderboard
Every model, every benchmark
A score of 90 means the model beat about 90% of tracked models on a typical test. The profile strip shows each category: darker means stronger.
| # | Model | Profile | ||||
|---|---|---|---|---|---|---|
1 | Claude Opus 5.5Anthropic | 95.9 | Consistent | 31 | $8.00 | |
2 | GPT-6 AstraOpenAI | 94.2 | Consistent | 34 | $20.00 | |
31 | Claude Fable 5.1Anthropic | 91.6 | Consistent | 33 | $20.00 | |
41 | Gemini 4 ArgonGoogle DeepMind | 91.2 | Consistent | 24 | $4.00 | |
5 | GPT-6.1 SolOpenAI | 90.9 | Consistent | 23 | $4.00 | |
6 | Claude Sonnet 5.5Anthropic | 90.6 | Consistent | 27 | $4.00 | |
7 | Muse WonderMeta AI | 87.9 | not enough data | 6 | — | |
8 | GPT-5.5 ProOpenAI | 86.6 | not enough data | 9 | $67.50 | |
9 | Claude Fable 5Anthropic | 86.2 | Consistent | 37 | $20.00 | |
10 | Claude Opus 5Anthropic | 84.9 | Consistent | 35 | $10.00 | |
11 | GPT-6 SolOpenAI | 84.7 | Consistent | 27 | $4.00 | |
12 | GPT-5.6 Sol ProOpenAI | 82.0 | not enough data | 7 | $8.00 | |
13 | GPT-5.6 SolOpenAI | 80.7 | Consistent | 37 | $8.00 | |
14 | MiMo-V2.6-ProXiaomi | 75.5 | Consistent | 12 | $0.54 | |
15 | Muse Spark 1.3Meta AI | 71.2 | Consistent | 23 | $2.00 | |
16 | GPT-5.4 ProOpenAI | 70.4 | Holds up | 13 | $67.50 | |
17 | GPT-5.5OpenAI | 69.7 | Consistent | 37 | $11.25 | |
18 | Claude Opus 4.8Anthropic | 68.7 | Consistent | 38 | $10.00 | |
19 | Gemini 3.8 FlashGoogle DeepMind | 64.1 | Consistent | 34 | $1.50 | |
20 | GPT-5.6 TerraOpenAI | 63.6 | Paper > real | 33 | $4.50 | |
216 | Claude Haiku 5.5Anthropic | 62.8 | Consistent | 24 | $0.20 | |
22 | Gemini 3.7 FlashGoogle DeepMind | 62.4 | Consistent | 27 | $1.50 | |
231 | Claude Opus 4.7Anthropic | 62.4 | Consistent | 34 | $10.00 | |
241 | Kimi K3Moonshot | 62.3 | Consistent | 34 | $4.09 | |
254 | Grok 4.6xAI | 62.0 | Consistent | 31 | $3.00 |
Models need at least 5 benchmarks across 3 categories to be ranked. Real use counts double in the overall score.
03Reality check
Great on paper, or great in use?
Some models are tuned to ace the well-known tests and then disappoint in real work. Each model's lab-test score sits next to its score on 11 tests that are hard to train for: blind human votes, brand-new tasks, private question sets and creators' own benchmarks.
- 1Claude Opus 5.5Consistent
- 2GPT-6 AstraConsistent
- 3Claude Fable 5.1Consistent
- 4Gemini 4 ArgonConsistent
- 5GPT-6.1 SolConsistent
- 6Claude Sonnet 5.5Consistent
- 9Claude Fable 5Consistent
- 10Claude Opus 5Consistent
- 11GPT-6 SolConsistent
- 13GPT-5.6 SolConsistent
- 14MiMo-V2.6-ProConsistent
- 15Muse Spark 1.3Consistent
- 16GPT-5.4 ProHolds up
- 17GPT-5.5Consistent
- 18Claude Opus 4.8Consistent
Flagged now: GPT-5.6 Terra, Step 5 Preview, GPT-5.6 Luna, DeepSeek V4 Pro, DeepSeek V4 Flash, Mistral Large 4 and 3 more. A flag is a reason to test it yourself, not a final verdict.
Under the microscope
Gemini 4 Argon: what creators and testers found
Lab tests 90, real-use tests 93. Hands-on reports below are linked and dated; they don't change the score.
- Worse than its benchmarks
Tested on 6 hard apps
Placed fourth overall (about 54% against Opus 5.5's 58%). Code quality 3.6/5 against Opus 5.5's 4.5/5, and it built backend features that didn't really work (keyboard sync, Fitbit sync, Docker downloads) while presenting them as finished. Best on the emulator and Pokémon game tasks, and far cheaper per app.
AICodeKing · YouTube · 14 Oct 2026
- Worse than its benchmarks
Noahbench, a personal coding benchmark
Called it weak at real-world code and suggested Google is aiming at other uses than the coding race.
Noah (Noahbench) · X · 13 Oct 2026
- Worse than its benchmarks
Staff question real-world performance
Staff with early access said it crushes standard benchmarks but struggles with real coding work and front-end website design. Surge AI's founder compared it to top SAT scores without real skills. Google disputes this.
Google employees, via Bloomberg · Report · 1 Oct 2026
- Mixed
Strong on question sets, weaker on real builds
Ranks 1st of 36 on the reasoning and data-analysis question sets, but 27th on design and layout work and 12th on coding.
LiveBench 2.0 suites · Data · 13 Oct 2026
- Better than expected
Box's complex-work evaluation
Box's CEO said it scored 14 percentage points higher on Box's evaluation of complex work.
Box (Aaron Levie) · Company eval
04Value
What you get for what you pay
05On deck
Waiting for more results
Too new or too few results to rank fairly. They join the board once enough benchmarks have tested them.
Mistral Medium 3.5
Mistral AI · 6 of 5 benchmarks
Muse Wonder
Meta · 6 of 5 benchmarks
Qwen3.8-Flash-Next
Alibaba · 4 of 5 benchmarks
Ring-2.6-1T
Ant Group · 4 of 5 benchmarks
Tencent Hy4 preview
Tencent · 4 of 5 benchmarks
Grok Build 0.1
xAI · 3 of 5 benchmarks
Mistral Small
Mistral AI · 3 of 5 benchmarks
Qwen3.8 2.4T A95B
Alibaba · 3 of 5 benchmarks
Frontier Board
Want to see every source?
Each benchmark, who runs it, what it tests, and the cost to run every model.