Thinking Machines AI Models: Benchmarks and Ranking

Rank	Model	Overall score	Artificial Analysis	LiveBench	Epoch AI	ARC-AGI	Scale SEAL	Vals AI	Terminal-Bench	EQ-Bench	SimpleBench	BenchLM
1	Claude Opus 5.5 Anthropic	92.5	#1	#2	—	#2	—	#1	—	—	#1	—
2	GPT-6 Astra OpenAI	92.2	#3	#4	#1	#1	#1	#4	#1	—	#4	#1
3	Claude Fable 5.1 Anthropic	91.7	#2	#1	#2	#5	#2	#2	#2	—	#2	#2
4	Claude Fable 5 Anthropic	90.6	#5	#3	#3	#6	—	#5	#4	#2	—	#4
5	Claude Opus 5 Anthropic	90.3	#4	#9	#4	#4	—	#3	#3	#1	#8	#3
6	Muse Spark 1.3 Meta	89.0	#6	#5	#15	—	—	#6	—	—	#7	—
7	GPT-5.6 Sol OpenAI	88.0	#8	#6	#6	#3	—	#7	#7	#9	#23	#5
8	GPT-5.5 Pro OpenAI	86.9	—	—	#5	#9	—	—	—	—	#10	—
9	GPT-6 Sol OpenAI	86.5	#7	#11	—	—	—	#8	—	—	#15	—
10	GPT-5.4 Pro OpenAI	85.2	—	—	#9	#13	—	—	—	—	—	—
11	MiMo-V2.6-Pro Xiaomi	85.2	#10	—	—	—	—	#16	—	—	—	—
12	DeepSeek V4.1 Flash DeepSeek	84.7	#22	#6	#28	—	—	#20	—	—	#21	—
13	Grok 4.7 xAI	84.6	#9	#19	—	—	—	#11	#6	—	—	—
14	GPT-5.5 OpenAI	84.4	#25	#8	#8	#8	—	#22	—	#4	#19	#12
15	Gemini 3.8 Flash Google	83.3	#19	#31	#13	#6	#4	#9	#11	—	#5	#6
16	Kimi K3 Moonshot AI	83.3	#15	#11	#12	#24	—	#21	—	#3	#33	#8
17	Qwen3.8 Max Alibaba (Qwen)	82.5	#11	#14	#26	—	—	#33	—	—	—	#9
18	Grok 4.6 xAI	82.4	#13	#15	#18	#20	—	#18	#10	—	#13	#11
19	Gemini 3.7 Flash Google	82.1	#23	#13	#11	#9	—	#17	#15	—	—	#14
19	GPT-5.4 OpenAI	82.1	—	#15	#14	#16	—	—	—	#7	—	#13
21	GPT-5.6 Terra OpenAI	82.1	#16	#18	#7	#12	—	#13	#9	#11	#53	#7
22	Muse Spark 1.2 Meta	82.0	#21	#15	#27	—	—	#23	—	—	#14	—
22	GLM-5.3 Z.ai (Zhipu)	82.0	#12	#28	#22	—	—	#24	#5	—	#22	—
24	Claude Opus 4.8 Anthropic	81.4	#17	#26	#10	#17	—	#10	#8	#6	#23	#10
25	Claude Opus 4.7 Anthropic	80.1	—	#25	#19	#15	#9	#25	—	#5	#26	—
26	Qwen3.8-Flash-Next Alibaba (Qwen)	78.7	#20	#26	—	—	—	—	—	—	—	—
27	Claude Sonnet 5 Anthropic	77.8	#26	#30	#20	—	—	#13	#13	#10	#34	#15
28	Gemini 3.1 Pro Google	77.3	#39	#22	#29	#14	#3	#41	—	#21	#9	—
29	DeepSeek V4 Pro DeepSeek	76.1	#29	#19	#24	#23	—	#32	—	#18	#51	—
30	Muse Spark 1.1 Meta	76.0	#33	#33	#33	—	—	#28	—	#8	—	#16
31	Grok 4.5 xAI	75.5	#24	#31	#34	#32	—	#34	#13	—	#18	#18
32	Claude Opus 4.6 Anthropic	74.8	—	#37	#25	#19	#10	—	—	#13	#20	#20
33	Muse Spark Meta	74.2	—	—	#39	—	#6	—	—	—	—	#27
34	Gemini 3.5 Flash Google	74.0	#36	#35	#30	#17	—	#30	—	#24	#11	#21
35	DeepSeek V4 Flash DeepSeek	74.0	#31	#38	#31	#22	—	#29	—	—	#31	—
36	GPT-5.6 Luna OpenAI	73.8	#27	#40	#21	#28	—	#12	#12	#19	#55	#17
37	Gemini 3.6 Flash Google	73.7	#32	#40	#32	#24	—	#26	—	—	—	#19
38	Qwen3.8 27B Alibaba (Qwen)	73.5	#33	#33	—	—	—	#36	—	—	—	—
39	GPT-6 Luna OpenAI	73.5	#27	#46	—	#29	—	#19	—	—	—	—
40	Gemini 3 Pro Google	70.4	—	—	#37	#37	#7	—	—	—	#12	#26
41	GPT-5.3 Codex OpenAI	70.0	—	—	#16	—	—	—	—	—	—	#24
42	GPT-5.2.11 OpenAI	69.9	—	#35	—	—	#12	—	—	—	—	—
43	GLM 5.3 Flash Z.ai (Zhipu)	69.7	#17	#48	#41	—	—	#37	—	—	—	—
44	GPT-5.2 Pro OpenAI	69.5	—	—	#23	#30	—	—	—	—	#41	—
45	Claude Sonnet 4.6 Anthropic	68.8	#38	#44	#38	#24	—	#35	—	#16	—	#30
46	GLM-5.2 Z.ai (Zhipu)	68.6	#33	#42	#43	#38	—	#30	—	#14	#38	#23
47	Qwen3.7 Max Alibaba (Qwen)	66.2	#40	#43	#35	—	—	#38	—	#23	—	#22
48	Qwen 3.6 Max Alibaba (Qwen)	65.8	—	—	#53	—	—	—	—	—	#25	—
49	GPT-5.2 OpenAI	65.2	—	—	#36	#31	—	—	—	—	#58	#25
50	Qwen 3.8 27B Alibaba (Qwen)	63.4	—	—	#52	—	—	—	—	—	#36	—

Rank	Provider	Overall score	Top model	Models
1	Anthropic	86.7	Claude Opus 5.5	27
2	OpenAI	85.2	GPT-6 Astra	54
3	Meta	77.7	Muse Spark 1.3	23
4	Google	74.4	Gemini 3.8 Flash	33
5	xAI	72.7	Grok 4.7	13
6	Xiaomi	70.1	MiMo-V2.6-Pro	4
7	Moonshot AI	69.8	Kimi K3	5
8	Alibaba (Qwen)	68.1	Qwen3.8 Max	46
9	DeepSeek	67.8	DeepSeek V4.1 Flash	22
10	Z.ai (Zhipu)	66.0	GLM-5.3	10
11	Tencent	53.3	—	2
12	Thinking Machines	42.0	Inkling	2
13	MiniMax	40.3	MiniMax-M3	5
14	Ant Group	32.7	—	5
15	NVIDIA	29.6	Nemotron 3 Ultra 550B A55B	9
16	Mistral AI	17.6	Mistral 3.5	26
17	Microsoft	16.7	Phi-4	6
18	Amazon	15.8	—	3
19	Inception	13.3	Mercury 2.5	1
20	Cohere	12.9	Command R	3

A provider’s result on each site is that of its best model there.

Performance vs price

Top left = cheap and strong. Click a point for details.

Best value

Overall score ÷ price

Top open-weight models

Models with public weights

Fastest models

Output tokens per second

By country

Top 50 models

Agreement between leaderboards

Rank correlation on shared models (1 = identical).

How the ranking works

Collect — Official data from the 11 leaderboards below is fetched every hour. Scores marked as estimates are not used.
Match — Different spellings of the same model are merged. For variants such as reasoning effort, each site’s best result is used.
Standardize — Each site’s main metric is standardized within that site, keeping score gaps rather than just ranks.
Equate — A factor model, bridged by models listed on several sites, corrects for each site’s difficulty and model mix. Noisier sites get less weight.
No penalty for missing data — Models are scored only from the sites that rated them. Uncertainty is shown as a 90% interval. Models rated on 2+ sites are ranked.
Arena — Arena (formerly LMArena) is used for domain rankings only, not for the overall score. New models are added automatically.

Thinking Machines AI models: benchmarks and ranking

More from Thinking Machines