Free reasoning models

Models with a reasoning (chain-of-thought) mode: better than plain instruct models on multi-step problems, math, and analysis, at the cost of more tokens.

58 models10 providersup to 1M contextbest: GLM 5.3Updated: September 26, 2026 at 07:55 PM UTC

Reasoning burns token quotas faster. Use Flash or non-reasoning modes for simple answers.

#	Model ↕	AA Index ↕	GPQA ↕	Context ↕	Size ↕	Free at ↕
1	GLM 5.3 Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	44.8	92%	1M	Frontier	Together AI, NVIDIA Build
2	Kimi K3 Moonshot AI VisionReasoningCodingTool useOpen weights	43.6	94%	1M	Frontier	NVIDIA Build, UnoRouter
3	GLM 5.3 Flash Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	41.8	91%	1M	Mid-size	OrcaRouter, NVIDIA Build, UnoRouter
4	Qwen 3.8 Flash Next Alibaba Cloud · New ReasoningCodingTool useOpen weights	39.8	92%	1M	Mid-size	UnoRouter
5	DeepSeek V4.1 Flash DeepSeek · New ReasoningCodingTool useOpen weights	39.5	—	—	Frontier	NVIDIA Build
6	DeepSeek V4 Flash DeepSeek ReasoningCodingTool useOpen weights	34.3	91%	1M	Frontier	OrcaRouter, UnoRouter
7	Qwen3.8 27B Alibaba Cloud · 27B ReasoningCodingTool useOpen weights	33.7	91%	262K	Mid-size	OpenRouter, Groq, UnoRouter
8	Gemini 3.6 Flash Google VisionReasoningCodingTool use	34.0	93%	1M	Mid-size	UnoRouter
9	GLM 5.2 Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	33.7	90%	1M	Frontier	Together AI
10	Inkling Small Thinking Machines Lab ReasoningOpen weights	27.8	90%	1M	Small	OpenRouter
11	Solar Pro 4 Upstage ReasoningTool use	28.2	89%	524K	Frontier	Nous Portal
12	Nemotron 3 Ultra NVIDIA · 550B · 55B ReasoningCodingTool useOpen weights	22.9	87%	1M	Frontier	OpenRouter, NVIDIA Build, UnoRouter
13	Inkling Thinking Machines Lab ReasoningOpen weights	25.0	87%	1M	Small	OpenRouter
14	Kimi K2.6 Moonshot AI ReasoningCodingTool useOpen weights	27.0	91%	—	Frontier	NVIDIA Build
15	Gemini 3.5 Flash Lite Google VisionReasoningCoding	22.2	84%	1M	Small	UnoRouter
16	GLM 4.7 Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	22.2	86%	203K	Frontier	Together AI
17	LongCat 2.0 Meituan ReasoningTool useOpen weights	19.1	78%	1M	Frontier	Nous Portal
18	Step 3.7 Flash StepFun ReasoningOpen weights	19.5	81%	262K	Small	Nous Portal
19	Qwen 3.5 35B Alibaba Cloud · 35B · 3B ReasoningCodingTool useOpen weights	19.3	85%	262K	Mid-size	Together AI
20	Qwen 3.6 35B Alibaba Cloud · 35B · 3B ReasoningCodingTool useOpen weights	18.2	84%	262K	Mid-size	Together AI
21	Qwen 3.5 122B Alibaba Cloud · 122B · 10B ReasoningCodingTool useOpen weights	17.7	83%	262K	Frontier	Together AI
22	MiniMax M2 MiniMax · 230B · 10B ReasoningCodingTool useOpen weights	18.6	78%	197K	Frontier	Together AI
23	Nemotron 3.5 Lightning NVIDIA · 30B · 3B ReasoningTool useOpen weights	12.9	74%	1M	Mid-size	OpenRouter, NVIDIA Build, UnoRouter
24	Nemotron 3 Super NVIDIA · 120B · 12B ReasoningCodingTool useOpen weights	12.8	80%	262K	Frontier	OpenRouter, NVIDIA Build, UnoRouter
25	Nemotron 3 Nano Omni NVIDIA · 30B · 3B VisionReasoningTool useOpen weights	10.3	47%	262K	Mid-size	+1OpenRouter, Together AI, NVIDIA Build, TokenRouter, UnoRouter
26	GLM 4.7 Flash Z.ai (Zhipu AI) · New ReasoningCodingTool useOpen weights	14.9	58%	128K	Mid-size	UnoRouter
27	Qwen 3.5 9B Alibaba Cloud · 9B ReasoningCodingOpen weights	13.7	81%	262K	Small	Together AI
28	Qwen3 VL 235B Alibaba Cloud · 235B · 22B VisionReasoningTool useOpen weights	13.4	77%	262K	Frontier	Together AI
29	GLM 4.7 Fp4 Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	—	—	203K	Frontier	Together AI
30	GLM 5 Fp4 Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	—	—	203K	Frontier	Together AI
31	MiniMax M2.5 Fp4 MiniMax ReasoningCodingTool useOpen weights	—	—	8K	Frontier	Together AI
32	Nvidia Nemotron 3 Super NVIDIA · 120B · 12B ReasoningCodingTool useOpen weights	12.8	80%	262K	Frontier	Together AI
33	MiniMax M1 80K MiniMax ReasoningTool useOpen weights	11.7	70%	1M	Frontier	Together AI
34	Qwen3 235B Alibaba Cloud · 235B · 22B ReasoningCodingTool useOpen weights	12.7	79%	262K	Frontier	Together AI
35	gpt-oss-120b OpenAI · 117B · 5.1B ReasoningCodingTool useOpen weights	11.6	78%	131K	Frontier	Groq, Cerebras
36	gpt-oss-20b OpenAI · 21B · 3.6B ReasoningTool useOpen weights	10.0	61%	131K	Mid-size	Groq, NVIDIA Build, UnoRouter
37	MiniMax M1 40K MiniMax ReasoningTool useOpen weights	10.0	68%	1M	Frontier	Together AI
38	Laguna S 2.1 Poolside ReasoningCoding	—	—	1M	Small	OpenRouter, Vercel AI Gateway, Nous Portal, UnoRouter
39	Qwen3 Next 80B Alibaba Cloud · 80B · 3B ReasoningCodingTool useOpen weights	11.2	76%	—	Frontier	Together AI
40	GLM 5.3 Flash Search Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	—	—	1M	Mid-size	UnoRouter
41	Qwen3 30B Alibaba Cloud · 30B · 3B ReasoningTool useOpen weights	9.8	71%	262K	Mid-size	Together AI
42	Qwen3 Coder 30B Alibaba Cloud · 30B · 3B ReasoningCodingTool useOpen weights	9.6	52%	262K	Frontier	Together AI
43	Laguna XS 2.1 Poolside ReasoningCoding	—	—	262K	Small	OpenRouter, NVIDIA Build, Nous Portal, UnoRouter
44	Nvidia Nemotron 3 Nano NVIDIA · 30B · 3B ReasoningTool useOpen weights	8.9	76%	262K	Mid-size	Together AI
45	Qwen3 4B Alibaba Cloud · 4B ReasoningOpen weights	8.8	67%	262K	Small	Together AI
46	QwQ 32B Alibaba Cloud · 32B · New ReasoningTool useOpen weights	9.5	59%	131K	Mid-size	UnoRouter
47	GLM 4.5 Flash Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	—	—	131K	Mid-size	UnoRouter
48	Qwen3 Alibaba Cloud · New ReasoningTool useOpen weights	—	—	25K	Mid-size	UnoRouter
49	Magistral Small Mistral AI ReasoningTool useOpen weights	8.6	66%	41K	Mid-size	Together AI
50	Qwen3 32B Alibaba Cloud · 32B ReasoningTool useOpen weights	8.6	67%	41K	Mid-size	Together AI
51	DeepSeek R1 Distill Qwen 32B DeepSeek · 32B · New ReasoningTool useOpen weights	8.4	62%	80K	Mid-size	UnoRouter
52	Qwen3 14B Alibaba Cloud · 14B ReasoningTool useOpen weights	8.2	60%	2K	Mid-size	Together AI
53	Qwen 3.5 2B Alibaba Cloud · 2B ReasoningCodingOpen weights	6.9	46%	262K	Small	Together AI
54	GLM 4 5v Z.ai (Zhipu AI) ReasoningCodingTool useOpen weights	7.6	68%	66K	Frontier	Together AI
55	Qwen3 8B Alibaba Cloud · 8B ReasoningOpen weights	7.3	59%	41K	Small	Together AI
56	Qwen 3.1 7B Alibaba Cloud · 1.7B ReasoningOpen weights	5.2	36%	41K	Small	Together AI
57	DeepSeek R1 Distill Qwen 7B DeepSeek · 7B ReasoningOpen weights	—	—	131K	Small	Together AI
58	Cosmos Reason2 8B NVIDIA · 8B VisionReasoning	—	—	—	Small	NVIDIA Build

Benchmarks: Artificial Analysis (September 24, 2026). A dash means the model is not on their leaderboard; we never estimate scores ourselves.

FAQ

What is the best free model for reasoning?

GLM 5.3 currently leads — free at Together AI, NVIDIA Build. Next come Kimi K3, GLM 5.3 Flash, Qwen 3.8 Flash Next.

FAQ

See also