Atlas

Benchmarks

DAXBench v2 (points score)

The points score DAXBench v2 ranks by: each fully correct task earns its complexity weight (basic 1.0, intermediate 1.3, advanced 1.6) plus a style bonus for best practices, performance and clarity, and the total is a percentage of the 147.36 available points; incorrect tasks earn nothing. Diagnostic, because the weights and the style bonus make it a composite rather than an accuracy; the primary daxbench-v2 counts fully correct tasks.

Models
82
Top score
95.9
Leader
GPT-6 Astra
Updated
—
Resolution
—
Saturation
—

No signal or difficulty estimate. Held out of the fit by design.

Organizations
82 dated models · 10 frontier modelsClosed weightsOpen weights
Data: DAXBenchatlas.village.dev

DAXBench v2 (points score) leaderboard

#ModelSettingScoreSource
1GPT-6 Astranot reported
95.9
2Claude Opus 5.5not reported
95.4
3Gemini 4 Carbonnot reported
92.8
4GPT-6.1 Terranot reported
91.9
5GPT-5.6 Solmedium
91.5
6GPT-6.1 Solnot reported
91.4
7Grok 4.7not reported
91.2
8Claude Fable 5.1not reported
90.7
9Claude Fable 5.2not reported
87.8
10Gemini 4 Argonnot reported
87.5
11GPT-5.5not reported
86.4
12Hy4not reported
85.7
13Gemini 3.8 Flashnot reported
85.6
14Claude Opus 5not reported
85.4
15Claude Sonnet 5.5not reported
84.6
16Grok 4.6not reported
83.9
17Kimi K3.1 Pro—
83.9
18GPT-6 Lunanot reported
83.0
19Muse Spark 1.3not reported
82.6
20Muse Wondernot reported
82.2
21Qwen3.8 Max (0902)not reported
81.7
22Gemini 3.1 Pro Previewnot reported
81.2
23MiMo-V2.6-Pronot reported
80.7
24Kimi K2.5not reported
79.7
25Claude Haiku 5.5not reported
79.4
26Kimi K3—
78.0
27Kimi K2.7 Codealways on
77.8
28GLM-5.3not reported
77.3
29GLM-5.3 Flashnot reported
75.8
30GPT-5.4 Minimedium
75.8
31DeepSeek V4.1 Flash 1017not reported
74.5
32DeepSeek V4.1 Flashnot reported
72.8
33Ling 3.1 Flashnot reported
72.7
34Gemini 2.5 Pro—
71.9
35GLM-4.5not reported
71.0
36Sonnet 5not reported
70.4
37GPT-5.1medium
70.4
38DeepSeek V4 Pro 0813not reported
66.3
39GLM-5not reported
66.0
40MiniMax M3—
64.2
41Gemini 3 Flash Previewnot reported
63.5
42Mistral Large 4 Previewnot reported
63.2
43GPT-5 Mini—
63.0
44MiniMax M3.1 Flash—
61.0
45Muse Glimmer 30Bnot reported
60.0
46Opus 4.5—
59.6
47Muse Embernot reported
59.4
48Qwen3.6 27Bnot reported
59.0
49MiMo-V2.6-Flashnot reported
58.9
50Sonnet 4not reported
58.4
51DeepSeek-R1always on
57.6
52Kimi K2 Instructnot supported
57.1
53Sonnet 4.5not reported
55.3
54Ripple-65B Previewunknown
55.0
55gpt-oss-120bnot reported
55.0
56Qwen3.5-27Bnot reported
53.8
57Gemma 4 26B A4B ITnot reported
53.2
58Mistral Large 2 (Instruct 2407)not supported
49.9
59Gemini 2.5 Flash—
49.9
60DeepSeek-V3.2not reported
49.8
61DeepSeek-V4-Flash-0731not reported
49.2
62Nemotron 3 Super 120B A12B—
48.5
63Qwen3.8 27Bx-high
48.2
64MiniMax M2always on
47.9
65MiniMax M2.5always on
46.7
66GPT-4o (2024-11-20)not supported
45.2
67Nemotron 3.5 Lightning 30B A3B (precision unspecified)not reported
44.0
68Llama 4 Maverick Instructnot supported
42.0
69Laguna S 2.1not reported
37.6
70Qwen3-30B-A3Bnot reported
37.0
71Haiku 4.5not reported
37.0
72GPT-4o Mininot supported
35.7
73LFM2.5-24B-A2Balways on
35.3
74gpt-oss-20bnot reported
34.4
75GLM 4.7 Flashnot reported
33.9
76Qwen2.5 Coder 32B Instructnot supported
31.9
77Llama 3.3 70B Instructnot supported
28.3
78Laguna XS 2.1not reported
27.5
79Gemma 3 27B ITnot supported
26.6
80Ministral 3 14B Instruct 2512not supported
23.2
81Llama 4 Scout Instructnot supported
21.4
82MiniMax M1always on
0.8

Related benchmarks

Most correlated

Benchmarks outside its family whose model rankings track it most reliably (Spearman ρ over at least 10 shared models).

Loading Atlas data…