Reasoning models
296 models that reason step by step before they answer, for math, code and multi-step problems.
/
| LFM2.5-2.6B (free)Liquid AIOpen weights · with termsJSON+2 | $0.50/1Mvia OpenRouter | 66K | $0.35/1M | $0.50/1M | — | 2K | |
 MiniMax M2.1MiniMaxOpen weights · with termsPrompt caching+3 | $0.864/1Mvia APIMart | 205K | $0.216/1M | $0.864/1M | — | 1K | |
 GLM-4.6VZhipu AIMITPrompt caching+4 | $0.411/1Mvia AIHubMix | 131K | $0.30/1M | $0.90/1M | 1376 | 1K | |
 ERNIE 5.0BaiduPrompt cachingReasoning+1 | $3.29/1Mvia AIHubMix | 119K | $0.822/1M | $3.29/1M | — | 1K | |
 Nemotron 3 Ultra 550B A55BNVIDIAReasoningTool calling | $2.20/1Mvia Deep Infra | 1M | Free/1M | Free/1M | — | 1K | |
| ERNIE 5.1BaiduPrompt cachingReasoning | $2.54/1Mvia AIHubMix | 128K | $0.563/1M | $2.54/1M | 1468 | 1K | |
 Fugu MaxSakana AIPrompt cachingJSON+3 | $6/1Mvia EmpirioLabs AI | 1M | $2/1M | $6/1M | — | 1K | |
 Mistral Small 4MistralPrompt cachingJSON+3 | $0.60/1Mvia EmpirioLabs AI | 262K | $0.15/1M | $0.60/1M | — | 930 | |
| Nemotron Lightning 3.5 30B A3BNVIDIAPrompt cachingJSON+2 | $0.20/1Mvia Requesty | 262K | $0.05/1M | $0.20/1M | — | 860 | |
| Nemotron 3 Nano Omni 30B TEENVIDIAJSONReasoning+1 | $0.24/1Mvia Requesty | 300K | $0.06/1M | $0.24/1M | — | 780 | |
| Nemotron 3 Nano OmniNVIDIAOpen weights · with termsReasoning+2 | Free/1Mvia Kenari | 256K | Free/1M | Free/1M | — | 670 | |
| Nemotron 3 Nano 30B A3BNVIDIAJSONReasoning+1 | $0.20/1Mvia Novita AI | 262K | $0.05/1M | $0.20/1M | — | 390 | |
 Hunyuan-A13B-InstructTencentOpen weights · with termsJSON+1 | $0.556/1Mvia SiliconFlow | 131K | $0.139/1M | $0.556/1M | — | 130 | |
 Doubao-Seed-CodeByteDancePrompt cachingReasoning+1 | $1.12/1Mvia ZenMux | 256K | $0.17/1M | $1.12/1M | — | 30 | |
 Aion 3.5Aion LabsPrompt cachingJSON+2 | $6/1Mvia OpenRouter | 262K | $3/1M | $6/1M | — | — | |
| Aion 3.5 MiniAion LabsPrompt cachingJSON+2 | $1.40/1Mvia OpenRouter | 262K | $0.70/1M | $1.40/1M | — | — | |
| Aion-3.0Aion LabsPrompt cachingJSON+2 | $6/1Mvia OpenRouter | 131K | $3/1M | $6/1M | — | — | |
| Aion-3.0-MiniAion LabsPrompt cachingJSON+2 | $1.40/1Mvia OpenRouter | 131K | $0.70/1M | $1.40/1M | — | — | |
 Arcee Trinity Large Thinking (Public Preview)Arcee AIOpen weights · permissivePrompt caching+2 | $0.80/1Mvia OpenRouter | 262K | $0.25/1M | $0.80/1M | — | — | |
 Beam 1.0 501B A23BReflectionPrompt cachingJSON+2 | $1/1Mvia OpenRouter | 1.05M | $0.25/1M | $1/1M | 1436 | — | |
| Claude 4.5 SonnetAnthropicPrompt cachingReasoning+2 | $15/1Mvia Replicate | 200K | $3/1M | $15/1M | — | — | |
| Claude Haiku 4.5 (latest)AnthropicPrompt cachingJSON+3 | $4/1Mvia APIMart | 200K | $1/1M | $5/1M | — | — | |
| Claude Opus 4AnthropicPrompt cachingJSON+3 | $75/1Mvia Atlas Cloud | 200K | $15/1M | $75/1M | 1377 | — | |
| Claude Opus 4.1AnthropicPrompt cachingReasoning+2 | $75/1Mvia AI/ML API | 200K | $15/1M | $75/1M | — | — | |
| Claude Opus 4.5 (latest)AnthropicPrompt cachingJSON+3 | $20/1Mvia APIMart | 200K | $5/1M | $25/1M | — | — | |
| Claude Opus 4.6AnthropicPrompt cachingJSON+3 | $20/1Mvia APIMart | 1M | $5/1M | $25/1M | 1546 | — | |
| Claude Opus 4.7AnthropicPrompt cachingJSON+3 | $20/1Mvia APIMart | 1M | $5/1M | $25/1M | 1558 | — | |
| Claude Opus 4.8AnthropicPrompt cachingJSON+3 | $20/1Mvia APIMart | 1M | $5/1M | $25/1M | 1556 | — | |
| Claude Sonnet 4AnthropicPrompt cachingJSON+3 | $15/1Mvia Atlas Cloud | 1M | $3/1M | $15/1M | 1351 | — | |
| Claude Sonnet 4.5 (latest)AnthropicPrompt cachingJSON+3 | $12/1Mvia APIMart | 1M | $3/1M | $15/1M | — | — | |
| Coding MiniMax M2.7 HighspeedMiniMaxPrompt cachingJSON+2 | $0.10/1Mvia AIHubMix | 205K | $0.10/1M | $0.10/1M | — | — | |
 Command A PlusCohereApache 2.0Prompt caching+4 | $1.50/1Mvia OpenRouter | 192K | $0.30/1M | $1.50/1M | — | — | |
 Deepseek 3.2DeepSeekMITPrompt caching+3 | $0.308/1Mvia APIMart | 164K | $0.206/1M | $0.308/1M | 1425 | — | |
| DeepSeek ChatDeepSeekPrompt cachingJSON+2 | $0.28/1Mvia Requesty | 164K | $0.14/1M | $0.28/1M | — | — | |
| DeepSeek R1 0528DeepSeekMITPrompt caching+3 | $1.23/1Mvia APIMart | 164K | $0.408/1M | $1.23/1M | — | — | |
| DeepSeek R1 Distill Llama 70BDeepSeekMITReasoning | $0.13/1Mvia Helicone | 32K | $0.03/1M | $0.13/1M | — | — | |
| DeepSeek ReasonerDeepSeekPrompt cachingJSON+2 | $0.28/1Mvia Requesty | 1.05M | $0.14/1M | $0.28/1M | — | — | |
| DeepSeek V3.1 TerminusDeepSeekMITJSON+2 | $0.95/1Mvia Hugging Face | 164K | $0.27/1M | $0.95/1M | 1419 | — | |
| DeepSeek V4 Flash 0424DeepSeekReasoningTool calling | $0.20/1Mvia Requesty | 1.05M | $0.10/1M | $0.20/1M | — | — | |
| DeepSeek V4 Flash Vision ExpDeepSeekMITPrompt caching+4 | $0.48/1Mvia CometAPI | 1.05M | $0.12/1M | $0.48/1M | — | — | |
| DeepSeek V4 ProDeepSeekMITPrompt caching+3 | $1.58/1Mvia CometAPI | 1M | $0.66/1M | $1.98/1M | — | — | |
| DeepSeek V4 Pro 0423DeepSeekPrompt cachingJSON+2 | $2.60/1Mvia Requesty | 1.05M | $1.30/1M | $2.60/1M | — | — | |
| DeepSeek-V3.1DeepSeekMITPrompt caching+3 | $0.798/1Mvia Poe | 164K | $0.212/1M | $0.798/1M | 1420 | — | |
| DeepSeek-V3.2-ExpDeepSeekMITJSON+2 | $0.308/1Mvia APIMart | 164K | $0.206/1M | $0.308/1M | 1425 | — | |
| Doubao Seed 1.6 LiteByteDancePrompt cachingJSON+3 | $0.656/1Mvia AIHubMix | 256K | $0.082/1M | $0.656/1M | — | — | |
| Doubao Seed 2.0 Code PreviewByteDancePrompt cachingReasoning+2 | $2.24/1Mvia ZenMux | 256K | $0.50/1M | $3/1M | — | — | |
| Doubao Seed 2.0 Lite 260428ByteDanceJSONReasoning+2 | $0.51/1Mvia ZenMux | 256K | $0.25/1M | $2/1M | — | — | |
| Doubao Seed 2.0 Mini 260428ByteDanceJSONReasoning+2 | $0.10/1Mvia Fal | 256K | $0.10/1M | $0.40/1M | — | — | |
| Doubao Seed 2.0 ProByteDancePrompt cachingJSON+3 | $2/1Mvia CometAPI | 256K | $0.50/1M | $3/1M | — | — | |
| Doubao Seed 2.1 ProByteDancePrompt cachingJSON+3 | $3.33/1Mvia CometAPI | 256K | $0.667/1M | $3.33/1M | — | — |
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