Multilingual

NOVA-63 leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on NOVA-63 is 59.1% by Qwen3.5 397B. 47 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 5.5	64.5%	estimated ± 1.6 pp, low confidence
2	Claude Mythos 5	63.9%	estimated ± 1.6 pp, low confidence
3	Claude Sonnet 5.5	63.2%	estimated ± 1.6 pp, low confidence
4	Claude Opus 5	62.9%	estimated ± 1.6 pp, low confidence
5	Claude Fable 5.1	62.8%	estimated ± 1.6 pp, low confidence
6	Claude Opus 4.8	61.0%	estimated ± 1.6 pp, low confidence
7	Hy4 preview	60.4%	estimated ± 1.6 pp, low confidence
8	Qwen3.8-Flash-Next	59.6%	estimated ± 1.6 pp, low confidence
9	Qwen3.8-Omni-Flash	59.4%	estimated ± 1.6 pp, low confidence
10	Composer 2.5	59.2%	estimated ± 1.6 pp, low confidence
11	Qwen3.5 397B	59.1%	measured
12	Ornith-1.5-397B	59.1%	estimated ± 1.6 pp, low confidence
13	Qwen3.7 Max	59.0%	measured
14	Qwen3.7 Plus	58.8%	measured
15	Ornith-1.0-397B	58.8%	estimated ± 1.6 pp, low confidence
16	Laguna S 2.1	58.6%	estimated ± 1.6 pp, low confidence
17	Claude Sonnet 5	58.5%	estimated ± 1.6 pp, low confidence
18	Grok 4.5	58.4%	estimated ± 1.6 pp, low confidence
19	Beam	58.4%	estimated ± 1.6 pp, low confidence
20	SWE-1.7	58.3%	estimated ± 1.6 pp, low confidence
21	Qwen3.6 Plus	57.9%	measured
22	Kimi K2.6	57.9%	estimated ± 1.6 pp, low confidence
23	MiniMax M2.7	57.8%	estimated ± 1.6 pp, low confidence
24	DeepSeek V4 Pro 0813	57.7%	estimated ± 1.6 pp, low confidence
25	dots3-note Preview	57.4%	estimated ± 1.6 pp, low confidence
26	Claude Opus 4.5	56.7%	measured
27	Nemotron 3 Ultra	56.7%	estimated ± 1.3 pp, low confidence
28	Composer 2	56.6%	estimated ± 1.6 pp, low confidence
29	DeepSeek V4 Flash 0731	56.4%	estimated ± 1.6 pp, low confidence
30	Qwen3.5-122B-A10B	56.3%	estimated ± 1.3 pp, low confidence
31	Qwen3.5-27B	56.3%	estimated ± 1.3 pp, low confidence
32	Kimi K2.5	56.0%	measured
33	Ling 3.0 Flash	56.0%	estimated ± 1.6 pp, low confidence
34	Qwen3.5-35B-A3B	55.7%	estimated ± 1.3 pp, low confidence
35	Ornith-1.5-35B-A3B	55.5%	estimated ± 1.6 pp, low confidence
36	Qwen3.6-27B	55.5%	estimated ± 1.6 pp, low confidence
37	GLM-5	55.1%	measured
38	Qwen3 235B 2507	54.9%	estimated ± 1.3 pp, low confidence
39	Ornith-1.0-35B	54.6%	estimated ± 1.6 pp, low confidence
40	Qwen3.6-35B-A3B	53.6%	estimated ± 1.6 pp, low confidence
41	Laguna M.1	51.6%	estimated ± 1.6 pp, low confidence
42	Laguna XS 2.1	51.6%	estimated ± 1.6 pp, low confidence
43	GPT-4.1	51.5%	estimated ± 1.3 pp, low confidence
44	DeepSeek V3 0324	50.3%	estimated ± 1.3 pp, low confidence
45	LongCat-Flash-Lite-Sparse	49.7%	estimated ± 1.6 pp, low confidence
46	Laguna XS.2	48.8%	estimated ± 1.6 pp, low confidence
47	Ornith-1.5-9B	47.0%	estimated ± 1.6 pp, low confidence
48	Ornith-1.0-9B	45.7%	estimated ± 1.6 pp, low confidence
49	GPT-4o	45.1%	estimated ± 1.3 pp, low confidence
50	Granite 4.2 30B	39.5%	estimated ± 1.6 pp, low confidence
51	Phi-4	38.4%	estimated ± 1.3 pp, low confidence
52	Nemotron 3.5 Lightning 30B A3B NVFP4	35.7%	estimated ± 1.6 pp, low confidence
53	Granite 4.2 8B	31.5%	estimated ± 1.6 pp, low confidence
54	LLaDA2.2-flash	26.8%	estimated ± 1.6 pp, low confidence

NOVA-63 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General