Reasoning models
296 models that reason step by step before they answer, for math, code and multi-step problems.
/
| Kimi K3.1 Pro Highspeed (Moonshot AI)Moonshot AIPrompt cachingReasoning+1 | $20/1Mvia Impossibl | 1.05M | $4/1M | $20/1M | — | 14K | |
| GPT OSS 20BOpenAIApache 2.0JSON+2 | $0.09/1Mvia OpenRouter | 131K | $0.018/1M | $0.09/1M | 1287 | 12K | |
| Mercury 2.5InceptionPrompt cachingJSON+2 | $0.15/1Mvia OpenRouter | 260K | $0.04/1M | $0.15/1M | — | 12K | |
 Qwen3.6 35B-A3BAlibabaApache 2.0Prompt caching+4 | $0.95/1Mvia Deep Infra | 262K | $0.375/1M | $2.25/1M | — | 11K | |
 Gemini 2.5 FlashGooglePrompt cachingJSON+3 | $2/1Mvia APIMart | 1.05M | — | $2.50/1M | 1417 | 11K | |
 Solar Pro 4UpstagePrompt cachingJSON+2 | $0.127/1Mvia AI/ML API | 524K | $0.032/1M | $0.127/1M | — | 11K | |
| Kimi K2.6Moonshot AIOpen weights · with termsPrompt caching+4 | $2.40/1Mvia APIYI | 262K | $0.903/1M | $3.75/1M | 1509 | 11K | |
 MiMo-V2.5XiaomiMITPrompt caching+4 | $0.224/1Mvia CometAPI | 1.05M | $0.112/1M | $0.224/1M | 1438 | 10K | |
| Qwen3.8 Omni FlashAlibabaPrompt cachingJSON+3 | $0.38/1Mvia AIHubMix | 1M | $0.113/1M | $0.382/1M | — | 10K | |
 DeepSeek V3DeepSeekOpen weightsReasoning+1 | $0.89/1Mvia Deep Infra | 64K | $0.32/1M | $0.89/1M | — | 9K | |
 Aion-2.0Aion LabsPrompt cachingJSON+2 | $1.60/1Mvia OpenRouter | 131K | $0.80/1M | $1.60/1M | — | 9K | |
| Mercury 3InceptionPrompt cachingJSON+2 | $0.35/1Mvia OpenRouter | 1M | $0.10/1M | $0.35/1M | — | 9K | |
 MiniMax M2.5MiniMaxOpen weights · with termsPrompt caching+3 | $0.95/1Mvia Venice AI | 205K | $0.27/1M | $0.95/1M | 1386 | 9K | |
 Gemma 4 26B A4BorcarouterJSONReasoning+2 | $2.90/1Mvia OrcaRouter | 262K | $0.25/1M | $2.90/1M | — | 8K | |
 GLM-4.7-FlashZhipu AIMITJSON+2 | $0.40/1Mvia Amazon Bedrock | 200K | $0.07/1M | $0.40/1M | 1351 | 7K | |
| GLM-5.1Zhipu AIMITPrompt caching+3 | $1.44/1Mvia APIMart | 200K | $1.40/1M | $4.40/1M | 1508 | 7K | |
| GLM 5.3 FlashxZhipu AIPrompt cachingJSON+3 | $0.987/1Mvia AIHubMix | 1.05M | $0.282/1M | $0.987/1M | — | 7K | |
| Solar Mini 4UpstagePrompt cachingJSON+2 | $0.20/1Mvia OpenRouter | 524K | $0.05/1M | $0.20/1M | — | 6K | |
 Ling 3.0 Flash (free)InclusionaiMITPrompt caching+2 | $0.063/1Mvia OpenRouter | 262K | $0.021/1M | $0.063/1M | — | 6K | |
| GLM-4.7Zhipu AIMITPrompt caching+3 | $1.10/1Mvia AIHubMix | 200K | $0.60/1M | $2.20/1M | 1435 | 6K | |
| GLM-4.5Zhipu AIMITPrompt caching+3 | $1.16/1Mvia ZenMux | 131K | $0.60/1M | $2.20/1M | 1430 | 6K | |
| GLM-4.6Zhipu AIMITPrompt caching+3 | $1.10/1Mvia AIHubMix | 200K | $0.60/1M | $2.20/1M | 1440 | 6K | |
| Gemma 4GoogleApache 2.0JSON+3 | $0.30/1Mvia FastRouter | 262K | $0.10/1M | $0.30/1M | — | 6K | |
| Kimi K2.7 Code HighSpeedMoonshot AIOpen weights · with termsPrompt caching+4 | $2.76/1Mvia Kenari | 262K | $0.903/1M | $3.75/1M | 1474 | 6K | |
| LFM2.5-24B-A2BLiquid AIOpen weights · with termsPrompt caching+3 | $0.70/1Mvia OpenRouter | 128K | $0.05/1M | $0.70/1M | — | 6K | |
| Qwen3-VL-8B-ThinkingAlibabaApache 2.0JSON+3 | $0.694/1Mvia SiliconFlow | 131K | $0.069/1M | $0.694/1M | — | 6K | |
| Kimi K2.8 PreviewMoonshot AIPrompt cachingReasoning+1 | $3.91/1Mvia CrazyRouter | 1.05M | $0.942/1M | $3.91/1M | — | 5K | |
| MiMo-V2.5-ProXiaomiMITPrompt caching+3 | $0.609/1Mvia OpenRouter | 1.05M | $0.304/1M | $0.609/1M | 1480 | 5K | |
| Qwen3 30B A3BAlibabaApache 2.0JSON+2 | $0.29/1Mvia Helicone | 41K | $0.08/1M | $0.29/1M | — | 5K | |
| GLM 5.3 PrimeZhipu AIPrompt cachingJSON+2 | $8.23/1Mvia AIHubMix | 1M | $2.35/1M | $8.23/1M | — | 5K | |
 Nemotron 3.5 Lightning 30B A3BNVIDIAReasoningTool calling | $0.16/1Mvia Deep Infra | 1M | Free/1M | Free/1M | — | 5K | |
| Qwen3.6 27BAlibabaApache 2.0JSON+3 | $2/1Mvia OpenRouter | 262K | $0.60/1M | $3.60/1M | — | 4K | |
| DeepSeek V4 Flash 0731DeepSeekMITPrompt caching+3 | $0.133/1Mvia OpenRouter | 1.05M | $0.044/1M | $0.133/1M | — | 4K | |
| LongCat-2.0MeituanMITPrompt caching+2 | $1.20/1Mvia Atlas Cloud | 1.05M | $0.30/1M | $1.20/1M | — | 4K | |
 Laguna S 2.1PoolsideReasoningTool calling | $0.18/1Mvia OpenRouter | 262K | $0.09/1M | $0.18/1M | — | 3K | |
 Mistral Medium 3.5MistralJSONReasoning+2 | $7.50/1Mvia Azure | 262K | $1.50/1M | $7.50/1M | 1421 | 3K | |
| Qwen3 14BAlibabaApache 2.0JSON+2 | $0.20/1Mvia Hugging Face | 41K | $0.07/1M | $0.20/1M | — | 3K | |
| Qwen3.5 397B-A17BAlibabaApache 2.0JSON+3 | $0.986/1Mvia AIHubMix | 262K | $0.172/1M | $1.03/1M | 1438 | 3K | |
| Qwen3.5 2BAlibabaApache 2.0JSON+3 | $0.10/1Mvia Requesty | 262K | $0.02/1M | $0.10/1M | — | 3K | |
| Mercury 2InceptionPrompt cachingJSON+2 | $0.75/1Mvia OpenRouter | 128K | $0.25/1M | $0.75/1M | 1355 | 3K | |
| Qwen3.7 MaxAlibabaPrompt cachingJSON+2 | $3.75/1Mvia DevPass (LLM Gateway) | 1M | $1.65/1M | $4.95/1M | 1515 | 3K | |
| Qwen3.5 35B-A3BAlibabaApache 2.0JSON+3 | $0.451/1Mvia AIHubMix | 262K | $0.25/1M | $2/1M | 1395 | 3K | |
| GLM-4.5-AirZhipu AIMITPrompt caching+3 | $0.291/1Mvia ZenMux | 131K | $0.20/1M | $1.10/1M | 1384 | 3K | |
| Qwen3.8 Max 0902AlibabaPrompt cachingJSON+3 | $4.11/1Mvia APIMart | 1M | $2/1M | $6/1M | — | 3K | |
| Gemma 4 26B A4B ITGoogleApache 2.0Prompt caching+4 | $0.225/1Mvia OpenRouter | 262K | $0.15/1M | $0.60/1M | — | 3K | |
| Qwen3 32BAlibabaApache 2.0JSON+2 | $0.25/1Mvia Hugging Face | 131K | $0.08/1M | $0.25/1M | 1340 | 2K | |
| Qwen3.7 PlusAlibabaPrompt cachingJSON+3 | $0.30/1Mvia CrazyRouter | 1M | $0.276/1M | $1.10/1M | 1455 | 2K | |
| Qwen3.5 27BAlibabaApache 2.0JSON+3 | $0.677/1Mvia AIHubMix | 262K | $0.30/1M | $2.40/1M | 1409 | 2K | |
| Qwen 3.8 Max PrimeAlibabaPrompt cachingJSON+3 | $12/1Mvia OpenRouter | 1M | $4/1M | $12/1M | — | 2K | |
| Nemotron 3 Super 120B A12BNVIDIAJSONReasoning+1 | $0.40/1Mvia Deep Infra | 262K | Free/1M | Free/1M | — | 2K |
Frequently asked questions
A model trained to work through a problem in intermediate steps, often called thinking, before it answers. It does better on math, coding and planning, at the cost of more tokens and slower replies.
The thinking steps are billed as output tokens, so one answer can use many times the tokens of a direct reply. Compare output prices, and check whether the model lets you limit or switch off its reasoning.
For short lookups, classification, extraction or chat, a regular model is faster and cheaper with little loss in quality. Keep reasoning for problems that really need several steps.
Can't find what you're looking for? Contact us