Lab comparison rankings.
A provider-level view of who is strongest across public benchmark coverage, best-category wins, open/proprietary mix, fastest matched models, and value leaders.
Labs covered
20
From public benchmark rows
Top lab score
Meta
Avg rank 20.2
Most models
Anthropic
Largest tracked model set
Open-weight depth
DeepSeek
9 open models
Provider view
Lab strength, coverage, and portfolio mix.
Average rank is computed across available Arena scores, so labs with fewer public rows should be read with coverage in mind.
Lab leaderboard
| Rank | Lab | Models | Avg rank | Best model | Best value | Fastest |
|---|---|---|---|---|---|---|
| #1 | Meta Coverage 46% | 70 open | 20.2Best #2 | Muse Wonder | Muse Spark 1.1 | Muse Spark Open |
| #2 | Anthropic Coverage 23% | 480 open | 30.6Best #1 | Claude Fable 5 | Claude Opus 4.6 | Claude Sonnet 5.5 |
| #3 | Google Coverage 31% | 252 open | 40.4Best #1 | Gemini 4 Argon | Gemini 4 Argon | Gemma 4 26b A4B |
| #4 | Stepfun Coverage 67% | 10 open | 58.3Best #36 | Step 5 Preview | Step 5 Preview | Step 5 Preview |
| #5 | Moonshot AI Coverage 37% | 65 open | 50.9Best #14 | Kimi K3 | Kimi K3 | Kimi K2.7 Code |
| #6 | Xiaomi Coverage 39% | 64 open | 58.8Best #17 | MiMo V2.6 Flash | MiMo V2.6 Pro | MiMo V2.6 Pro |
| #7 | xAI Coverage 26% | 150 open | 44Best #15 | Grok 4.20 Multi Agent Beta 0309 | Grok 4.5 | Grok 4.7 |
| #8 | Alibaba Coverage 36% | 146 open | 56.2Best #2 | Qwen3.8 Max | Qwen3.8 Max | Qwen3.5 122b A10B |
| #9 | OpenAI Coverage 27% | 470 open | 46Best #1 | GPT-5.6 Sol xHigh | GPT-6.1 Sol | GPT-6 Luna |
| #10 | Z.ai Coverage 32% | 98 open | 53.5Best #23 | GLM 5.3 | GLM 5.2 | GLM 5.3 Flash |
| #11 | Tencent Coverage 22% | 22 open | 50Best #18 | Hy4 Preview | Hy3 | No latency match |
| #12 | Perplexity Coverage 11% | 20 open | 37.5Best #37 | Ppl Sonar Pro High | No price match | No latency match |
| #13 | MiniMax Coverage 41% | 32 open | 76.7Best #38 | Minimax M3 | Minimax M3 | Minimax M3 |
| #14 | Diffbot Coverage 11% | 11 open | 42Best #42 | Diffbot Small Xl | No price match | No latency match |
| #15 | Mistral Coverage 37% | 32 open | 75.3Best #43 | Mistral Large 4 | Mistral Large 4 | Mistral Medium 3.5 |
| #16 | DeepSeek Coverage 22% | 99 open | 59.5Best #24 | DeepSeek V4.1 Flash | DeepSeek V4.1 Flash | DeepSeek V4.1 Flash |
| #17 | ByteDance Coverage 17% | 20 open | 55.7Best #49 | Dola Seed 2.0 Pro | No price match | No latency match |
| #18 | Thinky Coverage 56% | 22 open | 89.5Best #86 | Inkling Small | Inkling Small | Inkling Small |
| #19 | Baidu Coverage 14% | 40 open | 64.2Best #26 | ERNIE 5.1 | No price match | No latency match |
| #20 | Reflection Coverage 11% | 11 open | 79Best #79 | Beam 1.0 501b A23B | No price match | No latency match |
Meta
7 models, 0 open
Avg rank
20.2
Coverage
46%
Anthropic
48 models, 0 open
Avg rank
30.6
Coverage
23%
25 models, 2 open
Avg rank
40.4
Coverage
31%
Stepfun
1 models, 0 open
Avg rank
58.3
Coverage
67%
Moonshot AI
6 models, 5 open
Avg rank
50.9
Coverage
37%
Xiaomi
6 models, 4 open
Avg rank
58.8
Coverage
39%
xAI
15 models, 0 open
Avg rank
44
Coverage
26%
Alibaba
14 models, 6 open
Avg rank
56.2
Coverage
36%
OpenAI
47 models, 0 open
Avg rank
46
Coverage
27%
Z.ai
9 models, 8 open
Avg rank
53.5
Coverage
32%
Tencent
2 models, 2 open
Avg rank
50
Coverage
22%
Perplexity
2 models, 0 open
Avg rank
37.5
Coverage
11%
MiniMax
3 models, 2 open
Avg rank
76.7
Coverage
41%
Diffbot
1 models, 1 open
Avg rank
42
Coverage
11%
Mistral
3 models, 2 open
Avg rank
75.3
Coverage
37%
DeepSeek
9 models, 9 open
Avg rank
59.5
Coverage
22%
ByteDance
2 models, 0 open
Avg rank
55.7
Coverage
17%
Thinky
2 models, 2 open
Avg rank
89.5
Coverage
56%
Baidu
4 models, 0 open
Avg rank
64.2
Coverage
14%
Reflection
1 models, 1 open
Avg rank
79
Coverage
11%
Benchmark guide
What the scores mean.
A quick reading key for provider-level comparisons, coverage, average rank, and open-weight portfolio signals.
How is the lab score ranked?
Labs are ranked from available public Arena performance with a coverage adjustment, so broad benchmark coverage matters. A lab with one excellent score should not automatically outrank a lab with many strong model rows.
What does average rank mean?
Average rank is computed across the Arena rows available for that lab's tracked models. It is useful for comparing portfolio strength, but it should be read alongside model count and coverage.
Why do open-weight counts matter?
Open-weight counts show how much of a lab's tracked portfolio can plausibly be self-hosted or inspected outside a closed API. It is a portfolio signal, not a quality score by itself.
Why can labs with fewer models move around?
Small portfolios are more sensitive to one strong or weak model. The page keeps model count and coverage visible so lab comparisons are not reduced to a single rank number.