Atlas

Benchmarks

DAXBench v2 (syntax validity)

Share of DAXBench v2's 94 answers that were syntactically valid DAX, whether or not they returned the expected results. A diagnostic beside the primary daxbench-v2, which counts fully correct tasks.

Models
82
Top score
100.0
Leader
Gemini 3.8 Flash
Updated
—
Resolution
—
Saturation
—

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

Organizations
82 dated models · 8 frontier modelsClosed weightsOpen weights
Data: DAXBenchatlas.village.dev

DAXBench v2 (syntax validity) leaderboard

#ModelSettingScoreSource
1Gemini 3.8 Flashnot reported
100.0
2Grok 4.7not reported
100.0
3Gemini 4 Argonnot reported
98.9
4GPT-6 Astranot reported
98.9
5Kimi K2.7 Codealways on
98.9
6Claude Fable 5.1not reported
97.9
7Claude Haiku 5.5not reported
97.9
8Claude Opus 5not reported
97.9
9Claude Opus 5.5not reported
97.9
10Claude Sonnet 5.5not reported
97.9
11GPT-5.4 Minimedium
97.9
12Kimi K2.5not reported
97.9
13Qwen3.8 Max (0902)not reported
97.9
14Sonnet 5not reported
96.8
15Gemini 3.1 Pro Previewnot reported
96.8
16GPT-5.6 Solmedium
96.8
17GPT-6.1 Solnot reported
96.8
18GPT-6.1 Terranot reported
96.8
19Gemini 4 Carbonnot reported
95.7
20GPT-5.5not reported
95.7
21Kimi K3—
95.7
22Muse Spark 1.3not reported
95.7
23Qwen3.6 27Bnot reported
95.7
24GPT-5.1medium
94.7
25Sonnet 4not reported
93.6
26Gemini 2.5 Pro—
93.6
27Gemini 3 Flash Previewnot reported
93.6
28GPT-6 Lunanot reported
93.6
29Grok 4.6not reported
93.6
30Kimi K3.1 Pro—
93.6
31MiMo-V2.6-Flashnot reported
93.6
32DeepSeek-R1always on
92.6
33DeepSeek V4.1 Flash 1017not reported
92.6
34GLM-5.3not reported
92.6
35gpt-oss-120bnot reported
92.6
36Muse Wondernot reported
92.6
37Opus 4.5—
91.5
38DeepSeek V4.1 Flashnot reported
91.5
39Gemini 2.5 Flash—
91.5
40GLM-4.5not reported
91.5
41GLM-5.3 Flashnot reported
91.5
42Hy4not reported
91.5
43MiMo-V2.6-Pronot reported
91.5
44Gemma 4 26B A4B ITnot reported
90.4
45GPT-4o (2024-11-20)not supported
90.4
46GPT-5 Mini—
90.4
47Ling 3.1 Flashnot reported
90.4
48MiniMax M3—
90.4
49Mistral Large 2 (Instruct 2407)not supported
90.4
50Muse Glimmer 30Bnot reported
90.4
51GLM-5not reported
89.4
52Qwen3.5-27Bnot reported
89.4
53Claude Fable 5.2not reported
88.3
54Kimi K2 Instructnot supported
88.3
55Sonnet 4.5not reported
87.2
56GPT-4o Mininot supported
87.2
57Mistral Large 4 Previewnot reported
87.2
58Muse Embernot reported
87.2
59DeepSeek V4 Pro 0813not reported
86.2
60Qwen3.8 27Blow
85.1
61gpt-oss-20bnot reported
83.0
62MiniMax M2.5always on
83.0
63DeepSeek-V3.2not reported
81.9
64Laguna S 2.1not reported
80.9
65MiniMax M3.1 Flash—
80.9
66Qwen2.5 Coder 32B Instructnot supported
79.8
67Ripple-65B Previewunknown
79.8
68Nemotron 3 Super 120B A12B—
78.7
69Haiku 4.5not reported
77.7
70Laguna XS 2.1not reported
76.6
71Llama 4 Maverick Instructnot supported
76.6
72MiniMax M2always on
71.3
73Gemma 3 27B ITnot supported
69.1
74Llama 3.3 70B Instructnot supported
69.1
75GLM 4.7 Flashnot reported
68.1
76Nemotron 3.5 Lightning 30B A3B (precision unspecified)not reported
67.0
77Qwen3-30B-A3Bnot reported
67.0
78Llama 4 Scout Instructnot supported
63.8
79LFM2.5-24B-A2Balways on
61.7
80Ministral 3 14B Instruct 2512not supported
59.6
81DeepSeek-V4-Flash-0731not reported
58.5
82MiniMax M1always on
6.4

Related benchmarks

In the same family

Most correlated

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

Loading Atlas data…