Which visible model has the highest formal rank now?

Among the currently visible models, Claude Opus 4.5 has the highest formal position at global rank 1. Its Category Score is 63. 54 visible models meet the formal ranking rules. This result applies only to the run date and methodology version shown on the page.

Can I compare scores across different categories?

No. Each category uses different capability dimensions and evidence. A Category Score is comparable only inside the same leaderboard. Review the matching category for each task.

Are Estimated and Provisional models still useful?

They can help you find candidates, but their evidence is not complete enough for a formal rank. Review coverage and uncertainty, then test the model on a real task.

How often does the leaderboard update?

The leaderboard updates after a new completed score run is published. The current run date and methodology version appear above. LMSpeed does not promise a fixed daily or weekly schedule.

Is the number one model always best for me?

No. Your result also depends on speed, price, context length, tool support, region, and provider limits. Use the leaderboard to narrow the field, then run your own test.

Is a reasoning score the same as a math score?

No. Math is a separate category. The reasoning leaderboard also covers abstract logic, scientific causality, multi-step constraints, and evidence verification. A strong math score does not prove better results on every reasoning task.

GitHub

Sponsored byFusecodeEnterprise coding API for Claude Code, Codex, and model workflows.

LMSpeed

Best Models for Reasoning | LMSpeed

Rank	Model	LMSpeed score	Abstract logic		Scientific & causal reasoning						Multi-step constraints	Evidence integration & verification				Status	Evidence	Updated
Rank	Model	LMSpeed score	ARC-AGI-29 models	BenchLM Reasoning score2 models	GPQA186 models	HLE185 models	CritPt91 models	HLE w/o tools15 models	AA-GPQA Diamond5 models	AA-HLE5 models	MMLU-Pro125 models	AA-LCR90 models	LongBench v28 models	AI-Needle4 models	MRCR 1M3 models	Status	Evidence	Updated
Formally ranked models54
1		63±8.1	—	—

2	GPT-5.5OpenAI	61.5±8.8	85.0	—	92.6%	40.6%	27.1	41.4	—	—	—	74.3	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

3	Qwen3.7 MaxQwen	60.3±8.7	—	—	92.3%	38.1%	13.4	—	—	—	89.6%	69.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

4	Gemini 3.1 ProGoogle	59.9±8.8	77.1	—	—	—	17.7	45.4	94.1	44.7	—	72.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

5	Claude Opus 4.7 MaxAnthropic	58.6±8.8	75.8	—	—	—	12.0	46.9	91.4	39.6	—	70.3	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

6	Qwen3.7 PlusQwen	58.4±8.7	—	—	90.0%	33.4%	9.1	—	—	—	88.5%	65.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

7	Qwen3.6 PlusQwen	58.3±8.1	—	—	88.2%	25.7%	2.9	—	—	—	88.5%	69.7	62.0	68.3	—	Rated	3/4 dimensions · 7 families	Jul 14, 2026

8	GPT-5.1 CodexOpenAI	56.5±8.7	—	—	86.0%	23.4%	5.7	—	—	—	86.0%	67.3	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

9	Kimi K2.5MoonshotAI	56.3±8.3	—	—	87.9%	29.4%	3.1	—	—	—	87.1%	65.3	61.0	—	—	Rated	3/4 dimensions · 6 families	Jul 14, 2026

10	Gemini 2.5 ProGoogle	55.7±8.7	—	—	84.4%	21.1%	2.6	—	—	—	86.2%	66.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

11	DeepSeek V4 ProDeepSeek	55.3±8.4	—	—	71.7%	7.7%	10.0	—	—	—	82.9%	65.0	—	—	44.7	Rated	3/4 dimensions · 6 families	Jul 14, 2026

12	GLM-4.7Z.ai	55.3±8.7	—	—	85.9%	25.1%	1.7	—	—	—	85.6%	64.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

13	O3OpenAI	55.2±8.7	—	—	82.7%	20.0%	1.1	—	—	—	85.3%	69.3	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

14	Gemini 3.5 FlashGoogle	55.1±8.5	72.1	—	92.1%	39.9%	13.1	—	—	—	—	69.3	—	—	26.6	Rated	3/4 dimensions · 6 families	Jul 14, 2026

15	GPT-5.2OpenAI	55±6.5	52.9	—	86.4%	24.9%	11.6	—	—	—	85.9%	72.7	—	—	—	Rated	4/4 dimensions · 6 families	Jul 14, 2026

16	Qwen3.6 27BQwen	55±8.7	—	—	82.9%	13.6%	1.1	—	—	—	86.2%	68.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

17	Gemini 3 ProGoogle	54.6±6.5	31.1	—	90.8%	37.2%	9.1	—	—	—	89.8%	70.7	—	—	—	Rated	4/4 dimensions · 6 families	Jul 14, 2026

18	Qwen3.5-27BQwen	54.3±8.3	—	—	85.8%	22.2%	0.9	—	—	—	86.1%	67.3	60.6	—	—	Rated	3/4 dimensions · 6 families	Jul 14, 2026

19	Claude Opus 4.6Anthropic	54.3±8.7	—	—	89.6%	36.7%	2.8	40.0	—	—	82.0%	58.3	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

20	Qwen3.5-122B-A10BQwen	53.8±8.3	—	—	85.7%	23.4%	0.6	—	—	—	86.7%	66.7	60.2	—	—	Rated	3/4 dimensions · 6 families	Jul 14, 2026

21	Qwen3.5	53.5±8.1	—	—	45.6%	2.1%	1.7	—	—	—	87.8%	65.7	63.2	68.7	—	Rated	3/4 dimensions · 7 families	Jul 14, 2026

22	Qwen3.6 35B A3BQwen	53.2±8.7	—	—	81.7%	12.5%	0.3	—	—	—	85.2%	63.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

23	GPT-5.1OpenAI	52.8±8.7	—	—	64.3%	5.2%	4.9	—	—	—	80.1%	75.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

24	Gemini 3 FlashGoogle	52.8±8.7	—	—	81.2%	14.1%	1.4	—	—	—	88.2%	48.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

25	DeepSeek V4 FlashDeepSeek	52.7±8.4	—	—	71.6%	7.0%	3.4	—	—	—	83.0%	62.7	—	—	37.5	Rated	3/4 dimensions · 6 families	Jul 14, 2026

26	GPT-5OpenAI	52±8.7	—	—	68.6%	5.8%	0.0	—	—	—	82.0%	72.8	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

27	DeepSeek V3.2DeepSeek	51.7±8.7	—	—	87.1%	26.1%	0.9	—	—	—	86.3%	39.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

28	Qwen3.5-35B-A3BQwen	51.3±8.3	—	—	84.5%	19.7%	0.9	—	—	—	85.3%	62.7	59.0	—	—	Rated	3/4 dimensions · 6 families	Jul 14, 2026

29	O1OpenAI	51.2±8.7	—	—	74.7%	7.7%	0.3	—	—	—	84.1%	59.3	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

30	DeepSeek R1DeepSeek	51.2±8.7	—	—	70.8%	9.3%	1.4	—	—	—	84.4%	54.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

31	gpt-oss-120bOpenAI	50.6±8.7	—	—	78.2%	18.5%	1.1	—	—	—	80.8%	50.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

32	Claude Sonnet 4.6Anthropic	50.5±8.7	—	—	79.7%	10.8%	0.9	—	—	—	79.2%	57.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

33	DeepSeek V3.1DeepSeek	50.3±8.7	—	—	77.9%	13.0%	0.0	—	—	—	85.1%	45.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

34	Qwen3 MaxQwen	49.9±8.7	—	—	76.4%	11.1%	0.0	—	—	—	84.1%	46.7	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

35	Gemini 2.5 FlashGoogle	49.9±8.7	—	—	76.6%	7.8%	1.4	—	—	—	83.6%	45.9	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

36	GLM-5Z.ai	49.6±8.1	—	—	66.6%	7.2%	2.0	—	—	—	85.7%	63.3	60.8	63.3	—	Rated	3/4 dimensions · 7 families	Jul 14, 2026

37	Kimi K2MoonshotAI	49.5±8.7	—	—	76.6%	7.0%	0.0	—	—	—	82.4%	51.0	—	—	—	Rated	3/4 dimensions · 5 families	Jul 14, 2026

Estimated models — unranked46
—	GPT-5.6 SolOpenAI	62.3±10.8	—	—	93.1%	44.7%	32.3	—	—	—	—	73.7	—	—	—	Estimated	2/4 dimensions · 4 families	Jul 14, 2026
—		62.2±11.4	83.3	—	—
—		61±10.8	—	—	89.6%
—		61±10.8	—	—	91.5%
—		61±11.1	—	—	89.8%
—		60.4±10.8	—	—	92.0%
—		60.1±10.8	—	—	89.9%
—		59.8±10.8	—	—	92.9%
—		59.7±10.8	—	—	93.1%
—		59.6±10.8	—	—	87.1%
—		59.2±10.8	—	—	85.9%
—		57.7±10.8	—	—	89.6%
—		57.4±10.8	—	—	89.5%
—		57.4±11.1	—	—	83.0%
—		57.3±11.1	—	—	83.7%
—		57.2±10.8	—	—	88.8%
—		57±10.8	—	—	80.0%
—		57±10.8	—	—	88.5%
—		56.7±10.8	—	—	87.5%
—		56.6±10.8	—	—	89.0%
—		56.2±10.8	—	—	—
—		56.1±11.1	—	—	83.8%
—		56±11.1	—	—	83.6%
—		55.7±11.1	—	—	86.1%
—		55.7±10.8	—	—	78.8%
—		55.2±10.8	—	—	68.6%
—		55.1±10.8	—	—	87.4%
—		54.9±10.8	—	—	76.2%
—		54.7±10.8	—	—	83.9%
—		54.6±11.1	—	—	79.2%
—		54.4±11.1	—	—	79.7%
—		54.4±10.8	—	—	89.7%
—		54.2±11.1	—	—	79.0%
—		53.9±11.1	—	—	78.4%
—		53.9±10.8	—	—	82.8%
—		53.7±11.1	—	—	81.3%
—		53.6±10.8	—	—	80.9%
—		53.6±11.1	—	—	80.3%
—		53.2±11.1	—	—	78.2%
—		53.1±10.8	—	—	—
—		53.1±10.8	—	—	87.0%
—		52.8±11.3	53.3	—	91.1%
—		52.5±10.8	—	—	84.7%
—		52.5±11.1	—	—	77.7%
—		52±11.1	—	—	70.1%
—		51.6±11.1	—	—	77.3%

LMSpeed Best Models for Reasoning

Current answer

Available leaderboard data

What this leaderboard measures

Four capability dimensions

Abstract logic

Scientific & causal reasoning

Multi-step constraints

Evidence integration & verification

Reasoning tasks this page can help with

How to choose a model with this leaderboard

Check the rating status first

Review uncertainty and evidence

Test the real task last

Rating status guide

Rated

Estimated

Provisional

Benchmarks and evidence sources

How the reasoning model ranking is built

Leaderboard limits

Frequently asked questions