LuminaBench
Our data / Mathematics

Lumina Mathematics Index

By LuminaIndex designed by LuminaBench using attributed, publisher-run benchmarks. Lumina did not rerun those evaluations.

FrontierMath Tiers 1–3 and OTIS Mock AIME.

ExperimentalVerified 6 November 2026

Lumina Mathematics Index

LuminaBench
Show
Complete results · 2/2 · 20 of 62 models · best variant per model
03006009001,2001,094GPT-6.1 SolMax1,094GPT-6 AstraMax1,091Claude Opus 5.5Max1,091GPT-6.1 TerraMax1,090Claude Fable 5.1Max1,089GPT-6 SolMax1,088GPT-5.6 SolMax1,088Claude Sonnet 5.5Max1,085Claude Fable 5Max1,084Gemini 4 ArgonHigh1,083GPT-5.6 TerraMax1,082Claude Opus 5Max1,078Muse WonderMax1,077GPT-5.6 LunaMax1,074Claude Opus 4.8Max1,073Kimi K3.1 ProMax1,073GPT-6 LunaMax1,068GPT-5.4xHigh1,068Qwen 3.8 MaxxHigh1,067Muse Spark 1.3xHigh

Results table

RankModel / configurationLumina ScoreCoverage
1GPT-6.1 Sol
Max
1,0942/2 · Observed
1GPT-6 Astra
Max
1,0942/2 · Observed
3Claude Opus 5.5
Max
1,0912/2 · Observed
4GPT-6.1 Terra
Max
1,0912/2 · Observed
5Claude Fable 5.1
Max
1,0902/2 · Observed
6GPT-6 Sol
Max
1,0892/2 · Observed
7GPT-5.6 Sol
Max
1,0882/2 · Observed
8Claude Sonnet 5.5
Max
1,0882/2 · Observed
9Claude Fable 5
Max
1,0852/2 · Observed
10Gemini 4 Argon
High
1,0842/2 · Observed
11GPT-5.6 Terra
Max
1,0832/2 · Observed
12Claude Opus 5
Max
1,0822/2 · Observed
13Muse Wonder
Max
1,0782/2 · Observed
14GPT-5.6 Luna
Max
1,0772/2 · Observed
15Claude Opus 4.8
Max
1,0742/2 · Observed
16Kimi K3.1 Pro
Max
1,0732/2 · Observed
17GPT-6 Luna
Max
1,0732/2 · Observed
18GPT-5.4
xHigh
1,0682/2 · Observed
19Qwen 3.8 Max
xHigh
1,0682/2 · Observed
20Muse Spark 1.3
xHigh
1,0672/2 · Observed
21DeepSeek V4.1 Flash 1017
Max
1,0632/2 · Observed
22Kimi K3
Max
1,0622/2 · Observed
23Gemini 3.7 Flash
High
1,0612/2 · Observed
24Muse Spark Open
xHigh
1,0592/2 · Observed
25Gemini 3.8 Flash
High
1,0582/2 · Observed
26DeepSeek V4.1 Flash
Max
1,0562/2 · Observed
27Qwen3.8 Max (0902)
xHigh
1,0562/2 · Observed
28Grok 4.6
xHigh
1,0552/2 · Observed
29GPT-5.2
xHigh
1,0542/2 · Observed
30DeepSeek V4 Pro 0813
Max
1,0532/2 · Observed
31GLM-5.3
Max
1,0502/2 · Observed
32Gemini 3.5 Flash
High
1,0472/2 · Observed
33Claude Opus 4.7
Max
1,0462/2 · Observed
34Claude Opus 4.6
Max
1,0462/2 · Observed
35Mistral Large 4 Preview
High
1,0442/2 · Observed
36Grok 4.5
High
1,0422/2 · Observed
37Gemini 3.6 Flash
High
1,0402/2 · Observed
38DeepSeek V4 Flash 0731
Max
1,0382/2 · Observed
39Grok 4.7
xHigh
1,0362/2 · Observed
40GLM-5.3-Flash
Max
1,0352/2 · Observed
41Claude Sonnet 5
Max
1,0322/2 · Observed
42GPT-5
High
1,0322/2 · Observed
43GLM-5.2
Max
1,0312/2 · Observed
44DeepSeek-V4-Pro
Max
1,0242/2 · Observed
45GPT-5.4 Mini
xHigh
1,0232/2 · Observed
46Inkling-Small
xHigh
1,0182/2 · Observed
47Grok 4.3 Beta
High
1,0172/2 · Observed
48GPT-5 mini
High
1,0142/2 · Observed
49GPT-5.4 Nano
High
1,0132/2 · Observed
50GLM-5.2
Low
1,0122/2 · Observed
51Inkling
xHigh
9992/2 · Observed
52Qwen3.7-Plus
None
9892/2 · Observed
53o3
Medium
9882/2 · Observed
54Qwen3.5 397B-A17B
None
9882/2 · Observed
55GPT-5.6 Luna
Low
9832/2 · Observed
56Gemini 3.1 Flash-Lite
High
9802/2 · Observed
57o4-mini
Medium
9742/2 · Observed
58Qwen3.6 27B
None
9732/2 · Observed
59GPT-5 nano
High
9712/2 · Observed
60Gemini 3.5 Flash-Lite
High
9672/2 · Observed
61Qwen3.7 Flash
None
9662/2 · Observed
62Qwen 3.6 35B-A3B
None
9572/2 · Observed
62GPT-5.4 Nano
Low
9572/2 · Observed
64o1
High
9552/2 · Observed
65GPT-5.6 Luna
None
9492/2 · Observed
66o3
Low
9452/2 · Observed
67GLM-5.2
None
9402/2 · Observed
68o4-mini
Low
9382/2 · Observed
69Gemini 3.1 Flash-Lite
Low
9312/2 · Observed
70GPT-5
Minimal
9282/2 · Observed
71o1
Low
9222/2 · Observed
72Gemini 3.1 Flash-Lite
Minimal
9222/2 · Observed
73GPT-5 mini
Minimal
9222/2 · Observed
74GPT-5 nano
Low
9112/2 · Observed
75GPT-5.4 Nano
None
9092/2 · Observed
76o3-mini
Low
9052/2 · Observed
77GPT-5.4 Mini
None
9032/2 · Observed
78GPT-5 nano
Minimal
8922/2 · Observed
—Qwen 3.6 Plus
None
—1/2 · Source only
—Claude Haiku 4.5
Not specified
—1/2 · Source only
—Muse Spark
Not specified
—1/2 · Source only
—GPT-5.6 Sol
None
—1/2 · Source only
—GPT-5.5 Pro
xHigh
—1/2 · Source only
—Qwen3-Max
Not specified
—1/2 · Source only
—Claude Haiku 5.5
High
—1/2 · Source only
—Kimi K2 Thinking
Not specified
—1/2 · Source only
—GPT-5.4
High
—1/2 · Source only
—Claude Sonnet 4.5
Not specified
—1/2 · Source only
—Gemma 4 31B IT
Minimal
—1/2 · Source only
—GPT-5.6 Terra
None
—1/2 · Source only
—GPT-4 (Mar 2023)
Not specified
—1/2 · Source only
—Claude Opus 4.6
Not specified
—1/2 · Source only
—Qwen 3.6 Plus
Not specified
—1/2 · Source only
—GPT-3.5 Turbo (Jan 2024)
Not specified
—1/2 · Source only
—GPT-5.2
High
—1/2 · Source only
—Qwen 3.5 Plus (hosted 397B-A17B)
Not specified
—1/2 · Source only
—o1
Medium
—1/2 · Source only
—DeepSeek-R1-Distill-Qwen-14B
Not specified
—1/2 · Source only
—glm-4.7-flash
None
—1/2 · Source only
—GPT-5.5
xHigh
—1/2 · Source only
—Claude Opus 4.8
None
—1/2 · Source only
—Claude Opus 4.6
Not specified
—1/2 · Source only
—Gemini 3.1 Pro
Not specified
—1/2 · Source only
—Qwen 3.6 Flash
Not specified
—1/2 · Source only
—Llama 3.2 1B
Not specified
—1/2 · Source only
—Claude Haiku 4.5
Not specified
—1/2 · Source only
—GPT-5 Pro
High
—1/2 · Source only
—o3-mini
Medium
—1/2 · Source only
—Kimi K2.5
Not specified
—1/2 · Source only
—Claude Opus 4.1
Not specified
—1/2 · Source only
—GPT-4 Turbo (Apr 2024)
Not specified
—1/2 · Source only
—qwen3-4b-instruct-2507
Not specified
—1/2 · Source only
—granite-4.0-350m
Not specified
—1/2 · Source only
—GPT-5.4
None
—1/2 · Source only
—Kimi K3
Low
—1/2 · Source only
—Mistral Small 3
Not specified
—1/2 · Source only
—Qwen3.6 27B
Not specified
—1/2 · Source only
—Qwen3-32B
Not specified
—1/2 · Source only
—Qwen3-1.7B
None
—1/2 · Source only
—GLM-5.1
None
—1/2 · Source only
—Qwen3.7 Flash
Not specified
—1/2 · Source only
—granite-4.0-1b
Not specified
—1/2 · Source only
—Qwen3-30B-A3B
Not specified
—1/2 · Source only
—Claude Opus 4.7
xHigh
—1/2 · Source only
—Claude Sonnet 4.6
Medium
—1/2 · Source only
—Gemma 3 27B
Not specified
—1/2 · Source only
—Gemma 3 12B
Not specified
—1/2 · Source only
—Qwen3-8B
Not specified
—1/2 · Source only
—Qwen 3.6 Max (Preview)
Not specified
—1/2 · Source only
—GPT-5.4 Mini
Low
—1/2 · Source only
—Gemini 3 Flash
Not specified
—1/2 · Source only
—Claude Opus 4.5
Not specified
—1/2 · Source only
—Qwen3-14B
None
—1/2 · Source only
—Qwen 3.6 Flash
Not specified
—1/2 · Source only
—Kimi K2.7 Code
Not specified
—1/2 · Source only
—GPT-5.5 Instant
Not specified
—1/2 · Source only
—Claude Opus 5
Low
—1/2 · Source only
—Qwen3-Max
Not specified
—1/2 · Source only
—Kimi K3
High
—1/2 · Source only
—GPT-5.6 Terra
Low
—1/2 · Source only
—Claude Haiku 5.5
Max
—1/2 · Source only
—Claude Sonnet 4.5
Not specified
—1/2 · Source only
—GPT-3.5 Turbo (Jan 2024)
Not specified
—1/2 · Source only
—Qwen3.7-Plus
Not specified
—1/2 · Source only
—GPT-4o mini
Not specified
—1/2 · Source only
—gpt-oss-20b
Low
—1/2 · Source only
—Claude Opus 4.8
Low
—1/2 · Source only
—Mistral Small 3.1
Not specified
—1/2 · Source only
—Qwen3.6 27B
Not specified
—1/2 · Source only
—Gemini 2.5 Pro (Jun 2025)
Not specified
—1/2 · Source only
—GPT-5.1
Low
—1/2 · Source only
—Claude Opus 4.5
Not specified
—1/2 · Source only
—Qwen2.5-7B
Not specified
—1/2 · Source only
—Gemini 3 Flash
Not specified
—1/2 · Source only
—Claude Fable 5
High
—1/2 · Source only
—Grok 4.20
Not specified
—1/2 · Source only
—Kimi K2.7 Code
Not specified
—1/2 · Source only
—Qwen3.5-9B
Not specified
—1/2 · Source only
—Claude Opus 5
Not specified
—1/2 · Source only
—Claude Sonnet 4.6
Not specified
—1/2 · Source only
—GLM-5
Not specified
—1/2 · Source only
—GPT-4.1
Not specified
—1/2 · Source only
—Gemma 3 1B
Not specified
—1/2 · Source only
—Llama 3-8B
Not specified
—1/2 · Source only
—Llama 3.1-8B
Not specified
—1/2 · Source only
—Gemini 3.5 Flash
Low
—1/2 · Source only
—Gemini 3 Flash
High
—1/2 · Source only
—Qwen3.5-4B
None
—1/2 · Source only
—DeepSeek LLM 67B
Not specified
—1/2 · Source only
—Qwen 3.5 Flash (hosted 35B-A3B)
None
—1/2 · Source only
—GPT-5.2 Pro
xHigh
—1/2 · Source only
—Qwen3-30B-A3B
None
—1/2 · Source only
—DeepSeek-V4-Pro
None
—1/2 · Source only
—GPT-5.2
Medium
—1/2 · Source only
—Gemini 2.5 Pro (Jun 2025)
Not specified
—1/2 · Source only
—o4-mini
High
—1/2 · Source only
—Claude Opus 4.5
Not specified
—1/2 · Source only
—Claude Haiku 5.5
xHigh
—1/2 · Source only
—GPT-5.1
Medium
—1/2 · Source only
—Qwen3.5-35B-A3B
None
—1/2 · Source only
—Kimi K2.6
Not specified
—1/2 · Source only
—GPT-5.4
Medium
—1/2 · Source only
—GPT-5.1
None
—1/2 · Source only
—Gemini 3.5 Flash
Minimal
—1/2 · Source only
—QwQ-32B
Not specified
—1/2 · Source only
—Claude Sonnet 5
xHigh
—1/2 · Source only
—Qwen3.7-Max
Not specified
—1/2 · Source only
—GPT-5.5
None
—1/2 · Source only
—DeepSeek Reasoner
Not specified
—1/2 · Source only
—deepseek-r1-0528-qwen3-8b
Not specified
—1/2 · Source only
—Nemotron 3 Ultra
Not specified
—1/2 · Source only
—Claude Sonnet 4.6
Max
—1/2 · Source only
—Magistral Small 1.2
Not specified
—1/2 · Source only
—granite-4.0-micro
Not specified
—1/2 · Source only
—GPT-5.5 Instant
Not specified
—1/2 · Source only
—Gemini 3.1 Pro
High
—1/2 · Source only
—Qwen 3.6 35B-A3B
Not specified
—1/2 · Source only
—GPT-5
Low
—1/2 · Source only
—Claude Sonnet 4.6
High
—1/2 · Source only
—Qwen3.5-35B-A3B
Not specified
—1/2 · Source only
—Qwen 3.6 Plus
Not specified
—1/2 · Source only
—GLM-5.1
Not specified
—1/2 · Source only
—Qwen 3.5 Flash (hosted 35B-A3B)
Not specified
—1/2 · Source only
—GPT-5.1
High
—1/2 · Source only
—Gemini 3.1 Pro
Not specified
—1/2 · Source only
—Qwen3.7-Max
Not specified
—1/2 · Source only
—o3-mini
High
—1/2 · Source only
—GPT-5.4 Pro
xHigh
—1/2 · Source only
—Qwen3.5 397B-A17B
Not specified
—1/2 · Source only
—Qwen3-14B
Not specified
—1/2 · Source only
—Grok 4.20
Not specified
—1/2 · Source only
—DeepSeek-R1-Distill-Qwen-32B
Not specified
—1/2 · Source only
—MiniMax-M3
Not specified
—1/2 · Source only
—GPT-4o (Aug 2024)
Not specified
—1/2 · Source only
—DeepSeek-R1-Distill-Qwen-1.5B
Not specified
—1/2 · Source only
—Gemma 3 4B
Not specified
—1/2 · Source only
—MiniMax-M3
None
—1/2 · Source only
—gpt-oss-20b
Medium
—1/2 · Source only
—Qwen 3.6 Flash
None
—1/2 · Source only
—Claude Sonnet 4.5
Not specified
—1/2 · Source only
—GPT-5.2
None
—1/2 · Source only
—GPT-5.2
Low
—1/2 · Source only
—GPT-5.6 Sol
Low
—1/2 · Source only
—GPT-4 (Jun 2023)
Not specified
—1/2 · Source only
—Qwen 3.6 35B-A3B
Not specified
—1/2 · Source only
—Qwen3.7 Flash
Not specified
—1/2 · Source only
—GLM-4.7
Not specified
—1/2 · Source only
—Gemini 3 Pro
Not specified
—1/2 · Source only
—Grok 4.6
High
—1/2 · Source only
—Gemini 3.6 Flash
Minimal
—1/2 · Source only
—seed-oss-36b-instruct
Not specified
—1/2 · Source only
—Claude Sonnet 4.5
Not specified
—1/2 · Source only
—GPT-4.1 mini
Not specified
—1/2 · Source only
—Gemma 4 26B A4B
Minimal
—1/2 · Source only
—GPT-5.5 Pre Release
xHigh
—1/2 · Source only
—GPT-5 mini
Low
—1/2 · Source only
—Qwen3-8B
None
—1/2 · Source only
—gpt-oss-20b
High
—1/2 · Source only
—Qwen3-32B
None
—1/2 · Source only
—Claude Opus 4.5
Not specified
—1/2 · Source only
—GPT-5.5 Pro
xHigh
—1/2 · Source only
—GPT-5.5
Low
—1/2 · Source only
—GPT-5.4 Mini
High
—1/2 · Source only
—Muse Spark 1.3
Max
—1/2 · Source only
—DeepSeek Chat
Not specified
—1/2 · Source only
—Mistral Small 3.2
Not specified
—1/2 · Source only
—Mistral 7B v0.3
Not specified
—1/2 · Source only
—Kimi K2.6
Not specified
—1/2 · Source only
—Qwen3-30B-A3B-Thinking (Jul 2025)
Not specified
—1/2 · Source only
—o3
High
—1/2 · Source only
—glm-4.7-flash
Not specified
—1/2 · Source only
—Gemini 3.5 Flash-Lite
Minimal
—1/2 · Source only
—Claude Fable 5
Low
—1/2 · Source only
—DeepSeek-V4-Pro
High
—1/2 · Source only
—Qwen3.5-9B
None
—1/2 · Source only
—Gemini 3.5 Flash-Lite
Low
—1/2 · Source only
—Qwen3.5 397B-A17B
Not specified
—1/2 · Source only
—Gemini 3.6 Flash
Low
—1/2 · Source only
—Magistral Small 1.0
Not specified
—1/2 · Source only
—Qwen3-235B-A22B-Thinking (Jul 2025)
Not specified
—1/2 · Source only
—Qwen 3.5 Flash (hosted 35B-A3B)
Not specified
—1/2 · Source only
—GPT-5.4
Low
—1/2 · Source only
—Qwen3-30B-A3B-Instruct (Jul 2025)
Not specified
—1/2 · Source only
—GLM-5.1
Not specified
—1/2 · Source only
—gpt-oss-120b
High
—1/2 · Source only

Method & coverage

Equal-weight problem-solving scores under each publisher protocol. FrontierMath uses a Python scaffold; OTIS is a fixed 2024–25 contest cohort with ceiling effects. Both are evaluated by Epoch, so they are not independent evaluator confirmations. Tier 4 remains standalone: its different protocol and shared FrontierMath programme would reduce coverage and overweight one family. Formal proof and changing contest cohorts are not mixed into this definition. Two additional official Kimi K3.1 Pro Max observations use the retained Together route and the same protocols/conversions; they are scoped to Mathematics so existing Reasoning inputs remain frozen.

78 complete measured configurations; no missing-component estimates. Other observations remain source-only on their benchmark pages. Full-precision converted scores are weighted before rounding once for display. Costs, times, tokens and tool measurements always describe the selected benchmark workload.

Method: mathematics-index-observed-v2. Sources have recorded publication eligibility.