Models / API pricing
Compiled by Spring Prompt
LLM API pricing
What every model costs per million tokens, what you get for it, and when running an open model yourself would be cheaper. Prices are each developer's own list price where we have read it, otherwise the typical host on OpenRouter, with the cheapest host alongside.
- Models priced
- 203
- Best value
- 15
- no model is both cheaper and more capable
- Prices checked
- 7 Oct 2026
- Blended price
- 3 input tokens to 1 output
Every model's price
Dollars per million tokens. The price is the developer's own list price where we have read it; otherwise it's the typical (median) host on OpenRouter. "Cheapest host" is the cheapest route on OpenRouter for a typical mix, often a compressed (quantised) version of the model. Intelligence and tokens/s are from Artificial Analysis, which doesn't measure every model. Models with a shutdown date announced are left out.
| Model | Input | Output | Blended | Cheapest host blended | Intelligence | Our overall | Tokens/s |
|---|---|---|---|---|---|---|---|
| Mistral NemoMistral · open weights · typical host price | $0.023 | $0.030 | $0.025 | $0.021DekaLLM, FP8 · 6 hosts | – | – | – |
| Llama 3.1 8B InstructMeta · open weights · typical host price | $0.050 | $0.080 | $0.058 | $0.025DeepInfra, FP8 · 5 hosts | 6.9 | – | – |
| Mistral Small 3Mistral · open weights · typical host price | $0.050 | $0.080 | $0.058 | one host | 6.7 | – | – |
| gpt-oss-20bOpenAI · open weights · typical host price | $0.030 | $0.15 | $0.060 | $0.036Darkbloom, FP8 · 12 hosts | 10.0 | – | 249 |
| Nova Micro 1.0Amazon · typical host price | $0.035 | $0.14 | $0.061 | $0.061Amazon Bedrock · 2 hosts | 5.9 | – | 256 |
| Gemma 3 4BGoogle · open weights · typical host price | $0.050 | $0.10 | $0.062 | one host | 4.8 | – | – |
| Command R7B (12-2024)Cohere · typical host price | $0.037 | $0.15 | $0.066 | one host | – | – | – |
| Llama 3.2 1B InstructMeta · open weights · typical host price | $0.027 | $0.20 | $0.071 | one host | 4.8 | – | – |
| Gemma 3 12BGoogle · open weights · typical host price | $0.050 | $0.15 | $0.075 | one host | 3.8 | – | – |
| Nemotron 3 Nano 30B A3BNVIDIA · open weights · typical host price | $0.050 | $0.20 | $0.088 | $0.088Crusoe, FP8 · 4 hosts | 8.9 | – | 220 |
| Phi 4Microsoft · open weights · typical host price | $0.070 | $0.14 | $0.088 | one host | 5.9 | – | 42 |
| Ministral 3 3B 2512Mistral · open weights · developer price | $0.10 | $0.10 | $0.10 | $0.10Mistral · 3 hosts | 4.8 | – | 246 |
| Nova Lite 1.0Amazon · typical host price | $0.060 | $0.24 | $0.10 | $0.10Amazon Bedrock · 2 hosts | 6.7 | – | – |
| Granite 4.2 8BIBM · open weights · typical host price | $0.060 | $0.25 | $0.11 | $0.11DeepInfra, BF16 · 2 hosts | 11.1 | – | 51 |
| Qwen3.5-9BAlibaba · open weights · typical host price | $0.10 | $0.15 | $0.11 | $0.092Darkbloom, FP4 · 6 hosts | 13.3 | – | 89 |
| Qwen3.5-FlashAlibaba · typical host price | $0.065 | $0.26 | $0.11 | one host | – | – | – |
| Llama 3.2 3B InstructMeta · open weights · typical host price | $0.050 | $0.33 | $0.12 | $0.12Parasail, BF16 · 2 hosts | 5.7 | – | – |
| Qwen3 Coder 30B A3B InstructAlibaba · open weights · typical host price | $0.070 | $0.28 | $0.12 | $0.12Novita, FP8 · 4 hosts | 9.6 | – | – |
| Qwen2.5 7B InstructAlibaba · open weights · typical host price | $0.10 | $0.20 | $0.12 | one host | – | – | – |
| Reka Flash 3Rekaai · open weights · typical host price | $0.10 | $0.20 | $0.12 | one host | 5.6 | – | 95 |
| Mistral Small 3.2 24BMistral · open weights · typical host price | $0.094 | $0.25 | $0.13 | $0.11DeepInfra, FP8 · 4 hosts | 8.2 | – | – |
| GPT-5 NanoOpenAI · typical host price | $0.050 | $0.40 | $0.14 | $0.069OpenAI · 4 hosts | 13.0 | – | – |
| Gemma 3 27BGoogle · open weights · typical host price | $0.12 | $0.20 | $0.14 | $0.10DeepInfra, FP8 · 4 hosts | 4.9 | – | – |
| Qwen3 30B A3B Instruct 2507Alibaba · open weights · typical host price | $0.090 | $0.30 | $0.14 | $0.084StreamLake · 5 hosts | 7.5 | – | – |
| GLM 4.7 FlashZ.ai · open weights · typical host price | $0.060 | $0.40 | $0.15 | $0.15Venice, FP8 · 3 hosts | 14.9 | – | – |
| Ministral 3 8B 2512Mistral · open weights · developer price | $0.15 | $0.15 | $0.15 | $0.15Mistral · 3 hosts | 5.5 | – | 103 |
| Qwen3 14BAlibaba · open weights · typical host price | $0.12 | $0.24 | $0.15 | $0.13NextBit, INT4 · 3 hosts | 8.2 | – | – |
| Gemma 4 26B A4BGoogle · open weights · typical host price | $0.10 | $0.30 | $0.15 | $0.086Darkbloom · 13 hosts | 16.7 | – | 99 |
| Step 3.5 FlashStepfun · open weights · typical host price | $0.10 | $0.30 | $0.15 | one host | 16.6 | – | – |
| Solar Pro 4Upstage · typical host price | $0.090 | $0.36 | $0.16 | $0.16Upstage · 2 hosts | 28.2 | – | 100 |
| Nemotron 3 SuperNVIDIA · open weights · typical host price | $0.085 | $0.40 | $0.16 | $0.16DeepInfra, BF16 · 2 hosts | 12.8 | – | 150 |
| DeepSeek V4 Flash 0423DeepSeek · open weights · typical host price | $0.14 | $0.28 | $0.18 | $0.053StreamLake, FP8 · 16 hosts | 34.3 | – | – |
| GPT-4.1 NanoOpenAI · typical host price | $0.10 | $0.40 | $0.18 | $0.18Azure · 3 hosts | 7.8 | – | – |
| MiMo-V2.6-FlashXiaomi · open weights · developer price | $0.14 | $0.28 | $0.18 | $0.15Darkbloom, FP4 · 7 hosts | 37.9 | – | 52 |
| Seed-2.0-MiniBytedance · typical host price | $0.10 | $0.40 | $0.18 | one host | – | – | – |
| DeepSeek V4 Flash 0731DeepSeek · open weights · typical host price | $0.050 | $0.65 | $0.20 | $0.066StreamLake, FP8 · 26 hosts | – | – | – |
| GPT-6 LunaOpenAI · developer price | $0.10 | $0.50 | $0.20 | $0.10OpenAI · 7 hosts | 38.1 | 41 | 134 |
| Ministral 3 14B 2512Mistral · open weights · developer price | $0.20 | $0.20 | $0.20 | $0.20Mistral · 3 hosts | 6.0 | – | 91 |
| Gemma 4 31BGoogle · open weights · typical host price | $0.14 | $0.40 | $0.21 | $0.15DeepInfra, FP4 · 13 hosts | 14.7 | – | 44 |
| Qwen3 32BAlibaba · open weights · typical host price | $0.14 | $0.40 | $0.21 | $0.13DeepInfra, FP8 · 3 hosts | 8.6 | – | – |
| MiMo-V2.5Xiaomi · open weights · typical host price | $0.17 | $0.34 | $0.21 | $0.15GMICloud, FP8 · 5 hosts | 25.2 | – | – |
| Qwen3 30B A3BAlibaba · open weights · typical host price | $0.12 | $0.50 | $0.21 | $0.21DeepInfra, FP8 · 2 hosts | 7.6 | – | – |
| Qwen3.8 FlashAlibaba · open weights · typical host price | $0.15 | $0.47 | $0.23 | one host | – | – | – |
| GLM 5.3 FlashZ.ai · open weights · developer price | $0.15 | $0.50 | $0.24 | $0.12DeepInfra, FP4 · 33 hosts | 41.8 | – | 49 |
| Hy3Tencent · open weights · typical host price | $0.14 | $0.58 | $0.25 | $0.23DeepInfra, FP4 · 6 hosts | 25.3 | – | 89 |
| Command R (08-2024)Cohere · typical host price | $0.15 | $0.60 | $0.26 | one host | – | – | – |
| GPT-4o-mini (2024-07-18)OpenAI · typical host price | $0.15 | $0.60 | $0.26 | one host | 6.7 | – | – |
| Mistral Small 4Mistral · open weights · developer price | $0.15 | $0.60 | $0.26 | $0.26Mistral · 4 hosts | 11.3 | – | 164 |
| Qwen3 VL 30B A3B InstructAlibaba · open weights · typical host price | $0.15 | $0.60 | $0.26 | $0.23Alibaba · 4 hosts | 7.9 | – | – |
| Solar Pro 3Upstage · typical host price | $0.15 | $0.60 | $0.26 | one host | 7.8 | – | – |
| gpt-oss-120bOpenAI · open weights · typical host price | $0.15 | $0.60 | $0.26 | $0.065CoreWeave, FP4 · 23 hosts | 11.6 | – | 200 |
| Llama 4 ScoutMeta · open weights · typical host price | $0.18 | $0.59 | $0.28 | $0.15DeepInfra, FP8 · 3 hosts | 8.1 | – | 42 |
| Hy3 previewTencent · open weights · typical host price | $0.18 | $0.60 | $0.29 | one host | 22.7 | – | – |
| Qwen3 235B A22B Instruct 2507Alibaba · open weights · typical host price | $0.15 | $0.75 | $0.30 | $0.15GMICloud, FP8 · 10 hosts | 12.0 | – | – |
| DeepSeek V3.2 ExpDeepSeek · open weights · typical host price | $0.27 | $0.41 | $0.30 | one host | 16.6 | – | – |
| GLM 4.5 AirZ.ai · open weights · typical host price | $0.14 | $0.86 | $0.32 | $0.31Novita, BF16 · 3 hosts | 11.1 | – | – |
| Qwen3.6 35B A3BAlibaba · open weights · typical host price | $0.10 | $1.00 | $0.33 | $0.21Darkbloom, FP4 · 10 hosts | 18.2 | – | 140 |
| Qwen3 Next 80B A3B InstructAlibaba · open weights · typical host price | $0.10 | $1.10 | $0.35 | $0.27Alibaba · 5 hosts | 9.6 | – | 180 |
| Qwen3 Coder NextAlibaba · open weights · typical host price | $0.18 | $0.90 | $0.36 | $0.29Parasail, BF16 · 4 hosts | 9.2 | – | 104 |
| Qwen2.5 72B InstructAlibaba · open weights · typical host price | $0.36 | $0.40 | $0.37 | $0.37DeepInfra, FP8 · 2 hosts | 7.7 | – | – |
| Mercury 2Inception · typical host price | $0.25 | $0.75 | $0.38 | one host | 13.8 | – | – |
| Trinity Large ThinkingArcee Ai · open weights · typical host price | $0.25 | $0.80 | $0.39 | one host | 10.8 | – | 344 |
| Qwen-PlusAlibaba · typical host price | $0.26 | $0.78 | $0.39 | one host | – | – | – |
| Llama 3.1 70B InstructMeta · open weights · typical host price | $0.40 | $0.40 | $0.40 | $0.40DeepInfra, FP8 · 2 hosts | 6.6 | – | – |
| Mistral Small 3.1 24BMistral · open weights · typical host price | $0.35 | $0.56 | $0.40 | one host | 7.1 | – | – |
| Llama 4 MaverickMeta · open weights · typical host price | $0.27 | $0.85 | $0.42 | $0.30DigitalOcean · 4 hosts | 10.0 | – | 42 |
| Qwen3.6 FlashAlibaba · typical host price | $0.19 | $1.12 | $0.42 | one host | – | – | – |
| DeepSeek V3 0324DeepSeek · open weights · typical host price | $0.25 | $1.00 | $0.44 | $0.44SiliconFlow, FP8 · 2 hosts | 9.7 | – | – |
| Qwen3.5-35B-A3BAlibaba · open weights · typical host price | $0.16 | $1.30 | $0.45 | $0.25Darkbloom, FP4 · 7 hosts | 19.3 | – | – |
| GLM 4.6VZ.ai · open weights · typical host price | $0.30 | $0.90 | $0.45 | $0.45Novita, BF16 · 2 hosts | 11.2 | – | – |
| GPT-5.6 LunaOpenAI · developer price | $0.20 | $1.20 | $0.45 | $0.23OpenAI · 7 hosts | 37.3 | – | – |
| DeepSeek V3DeepSeek · open weights · typical host price | $0.26 | $1.03 | $0.45 | $0.45StreamLake · 2 hosts | 8.5 | – | – |
| DeepSeek V3.1 TerminusDeepSeek · open weights · typical host price | $0.27 | $1.00 | $0.45 | $0.45SiliconFlow, FP8 · 2 hosts | 14.8 | – | – |
| GPT-5.4 NanoOpenAI · developer price | $0.20 | $1.25 | $0.46 | $0.23OpenAI · 4 hosts | 20.7 | – | – |
| DeepSeek V3.2DeepSeek · open weights · typical host price | $0.30 | $0.96 | $0.46 | $0.23GMICloud, FP8 · 13 hosts | 21.5 | – | – |
| DeepSeek V4.1 FlashDeepSeek · open weights · developer priceHalf price off-peak ($0.15 / $0.60) | $0.30 | $1.20 | $0.52 | $0.11Decart, FP4 · 29 hosts | 39.5 | 58 | 216 |
| MiniMax M2MiniMax · open weights · typical host price | $0.30 | $1.20 | $0.52 | $0.45Minimax, FP8 · 3 hosts | 18.6 | – | – |
| MiniMax M2.1MiniMax · open weights · typical host price | $0.30 | $1.20 | $0.52 | $0.52Novita, FP8 · 3 hosts | 20.9 | – | – |
| MiniMax M2.5MiniMax · open weights · typical host price | $0.30 | $1.20 | $0.52 | $0.44Venice · 9 hosts | 22.8 | – | – |
| MiniMax M2.7MiniMax · open weights · developer price | $0.30 | $1.20 | $0.52 | $0.37GMICloud, FP8 · 7 hosts | 22.8 | – | – |
| MiniMax M3MiniMax · open weights · developer priceShown as a permanent 50% off $0.60 / $2.40 | $0.30 | $1.20 | $0.52 | $0.41CoreWeave, FP4 · 13 hosts | 29.2 | – | 81 |
| Muse Glimmer 30BMeta · open weights · typical host price | $0.30 | $1.20 | $0.52 | $0.50Phala · 3 hosts | – | – | – |
| MiMo-V2.5-ProXiaomi · open weights · typical host price | $0.43 | $0.87 | $0.54 | $0.38GMICloud, BF16 · 6 hosts | 26.0 | – | – |
| MiMo-V2.6-ProXiaomi · open weights · developer price | $0.43 | $0.87 | $0.54 | $0.54DeepInfra, FP8 · 4 hosts | 46.3 | – | 40 |
| Qwen3.7 PlusAlibaba · typical host price | $0.32 | $1.28 | $0.56 | one host | 25.2 | – | 56 |
| Gemini 3.1 Flash LiteGoogle · typical host price | $0.25 | $1.50 | $0.56 | $0.28Google · 8 hosts | – | – | – |
| Gemini 3.1 Flash Lite PreviewGoogle · typical host price | $0.25 | $1.50 | $0.56 | $0.28Google AI Studio · 3 hosts | 15.6 | – | – |
| Qwen3.5 Plus 2026-02-15Alibaba · typical host price | $0.26 | $1.56 | $0.58 | one host | – | – | – |
| Qwen3 VL 235B A22B InstructAlibaba · open weights · typical host price | $0.30 | $1.50 | $0.60 | $0.37DeepInfra, FP8 · 5 hosts | 9.9 | – | – |
| Inkling SmallThinkingmachines · open weights · typical host price | $0.45 | $1.20 | $0.64 | one host | 25.7 | – | 173 |
| Qwen3 Coder 480B A35BAlibaba · open weights · typical host price | $0.35 | $1.50 | $0.64 | $0.47DeepInfra, FP4 · 5 hosts | 11.9 | – | – |
| Llama 3.3 70B InstructMeta · open weights · typical host price | $0.59 | $0.79 | $0.64 | $0.16DeepInfra, FP8 · 11 hosts | 7.7 | – | 80 |
| Gemma 2 27BGoogle · open weights · typical host price | $0.65 | $0.65 | $0.65 | one host | – | – | – |
| DeepSeek V4 Flash Vision ExpDeepSeek · open weights · typical host price | $0.44 | $1.32 | $0.66 | $0.32DeepInfra, FP8 · 4 hosts | 34.8 | – | 217 |
| GPT-5 MiniOpenAI · typical host price | $0.25 | $2.00 | $0.69 | $0.34OpenAI · 4 hosts | 20.6 | – | – |
| GPT-5.1-Codex-MiniOpenAI · typical host price | $0.25 | $2.00 | $0.69 | one host | 20.4 | – | – |
| GPT-4.1 MiniOpenAI · typical host price | $0.40 | $1.60 | $0.70 | $0.70Azure · 3 hosts | 10.2 | – | – |
| Qwen3.5-122B-A10BAlibaba · open weights · typical host price | $0.26 | $2.08 | $0.72 | $0.72SiliconFlow, FP8 · 4 hosts | 17.7 | – | 147 |
| Qwen3.6 PlusAlibaba · typical host price | $0.33 | $1.95 | $0.73 | one host | 27.0 | – | – |
| Qwen3.5-27BAlibaba · open weights · typical host price | $0.27 | $2.16 | $0.74 | $0.54Alibaba · 6 hosts | 22.9 | – | – |
| Qwen2.5 Coder 32B InstructAlibaba · open weights · typical host price | $0.66 | $1.00 | $0.74 | one host | 6.7 | – | – |
| Mistral Large 3 2512Mistral · developer price | $0.50 | $1.50 | $0.75 | $0.75Mistral · 2 hosts | 9.3 | – | 80 |
| Devstral 2 2512Mistral · open weights · typical host price | $0.40 | $2.00 | $0.80 | one host | 8.6 | – | – |
| Mistral Medium 3Mistral · typical host price | $0.40 | $2.00 | $0.80 | $0.80Mistral · 3 hosts | 9.0 | – | – |
| Mistral Medium 3.1Mistral · typical host price | $0.40 | $2.00 | $0.80 | $0.80Mistral · 3 hosts | 9.2 | – | – |
| DeepSeek V3.1DeepSeek · open weights · typical host price | $0.55 | $1.65 | $0.82 | $0.42DeepInfra, FP4 · 5 hosts | 13.7 | – | – |
| Gemini 3.5 Flash LiteGoogle · developer price | $0.30 | $2.50 | $0.85 | $0.42Google · 8 hosts | 22.2 | 27 | 330 |
| Nova 2 LiteAmazon · typical host price | $0.30 | $2.50 | $0.85 | $0.85Amazon Bedrock · 2 hosts | 13.4 | – | 210 |
| MiniMax M1MiniMax · typical host price | $0.40 | $2.20 | $0.85 | $0.85Minimax · 2 hosts | – | – | – |
| Qwen2.5 VL 72B InstructAlibaba · open weights · typical host price | $0.80 | $1.00 | $0.85 | one host | – | – | – |
| GLM 4.6Z.ai · open weights · typical host price | $0.50 | $2.00 | $0.88 | $0.76Venice, FP4 · 4 hosts | 18.5 | – | – |
| GLM 4.5VZ.ai · open weights · typical host price | $0.60 | $1.80 | $0.90 | $0.90Novita, FP8 · 2 hosts | 7.6 | – | – |
| R1 0528DeepSeek · open weights · typical host price | $0.50 | $2.18 | $0.92 | $0.91DeepInfra, FP4 · 4 hosts | 13.1 | – | – |
| Kimi K2 0711Moonshot AI · open weights · typical host price | $0.57 | $2.30 | $1.00 | one host | 12.7 | – | – |
| Qwen3.6 27BAlibaba · open weights · typical host price | $0.30 | $3.20 | $1.02 | $0.72Chutes, FP8 · 6 hosts | 21.4 | – | – |
| Mistral Large 4Mistral · developer priceLaunch price, half off until about 20 Oct; list $1.36 / $4.18 | $0.68 | $2.09 | $1.03 | one host | 38.4 | 31 | 106 |
| Kimi K2 0905Moonshot AI · open weights · typical host price | $0.60 | $2.50 | $1.07 | one host | 15.3 | – | – |
| Kimi K2 ThinkingMoonshot AI · open weights · typical host price | $0.60 | $2.50 | $1.07 | $1.07Google · 2 hosts | 22.0 | – | – |
| Gemini 3 Flash PreviewGoogle · typical host price | $0.50 | $3.00 | $1.12 | $0.56Google · 6 hosts | – | – | – |
| Qwen3.8 27BAlibaba · open weights · developer price | $0.50 | $3.00 | $1.12 | $0.58Phala · 18 hosts | 33.7 | – | 52 |
| Kimi K2.5Moonshot AI · open weights · typical host price | $0.57 | $2.85 | $1.14 | $0.90SiliconFlow, INT4 · 5 hosts | 23.5 | – | – |
| R1DeepSeek · open weights · typical host price | $0.70 | $2.50 | $1.15 | one host | 11.4 | – | – |
| Grok Build 0.1xAI · developer price | $1.00 | $2.00 | $1.25 | $1.25xAI · 4 hosts | 27.2 | – | – |
| Hy4 previewTencent · open weights · typical host price | $0.83 | $2.50 | $1.25 | $1.25DeepInfra, FP8 · 4 hosts | – | – | – |
| Qwen3.5 397B A17BAlibaba · open weights · typical host price | $0.55 | $3.50 | $1.29 | $0.88Alibaba · 10 hosts | 21.4 | – | 90 |
| GLM 5Z.ai · open weights · typical host price | $0.95 | $2.55 | $1.35 | $0.93StreamLake, FP8 · 8 hosts | 27.9 | – | – |
| Nova Pro 1.0Amazon · typical host price | $0.80 | $3.20 | $1.40 | $1.40Amazon Bedrock · 2 hosts | 7.0 | – | – |
| Gemini 3.6 FlashGoogle · developer priceIntroductory price; $1.50 / $7.50 from 1 Jan 2027 | $0.75 | $3.75 | $1.50 | $0.75Google · 7 hosts | 34.0 | – | – |
| Gemini 3.7 FlashGoogle · developer priceIntroductory price; $1.50 / $7.50 from 1 Jan 2027 | $0.75 | $3.75 | $1.50 | $0.75Google · 6 hosts | 39.6 | – | – |
| Gemini 3.8 FlashGoogle · developer priceIntroductory price; $1.50 / $7.50 from 1 Jan 2027 | $0.75 | $3.75 | $1.50 | $0.75Google AI Studio · 6 hosts | 40.9 | 61 | 187 |
| Grok 4.20xAI · typical host price | $1.25 | $2.50 | $1.56 | $1.56xAI · 4 hosts | 25.7 | – | – |
| Grok 4.3xAI · developer price | $1.25 | $2.50 | $1.56 | $1.56xAI · 4 hosts | 24.9 | – | – |
| GPT-5.4 MiniOpenAI · developer price | $0.75 | $4.50 | $1.69 | $0.84OpenAI · 5 hosts | 24.1 | – | – |
| Kimi K2.6Moonshot AI · open weights · developer price | $0.95 | $4.00 | $1.71 | $0.96Inceptron, INT4 · 18 hosts | 27.0 | – | – |
| Kimi K2.7 CodeMoonshot AI · open weights · developer price | $0.95 | $4.00 | $1.71 | $1.28StreamLake · 12 hosts | 25.8 | – | 83 |
| InklingThinkingmachines · open weights · typical host price | $0.95 | $4.05 | $1.72 | $1.72DeepInfra, FP8 · 2 hosts | 25.0 | – | 166 |
| GLM 5 TurboZ.ai · typical host price | $1.20 | $4.00 | $1.90 | one host | 26.6 | – | – |
| GLM 5V TurboZ.ai · typical host price | $1.20 | $4.00 | $1.90 | one host | 23.5 | 12 | – |
| o3 MiniOpenAI · typical host price | $1.10 | $4.40 | $1.93 | one host | 12.5 | – | – |
| o3 Mini HighOpenAI · typical host price | $1.10 | $4.40 | $1.93 | one host | 11.0 | – | – |
| o4 MiniOpenAI · typical host price | $1.10 | $4.40 | $1.93 | one host | 16.7 | – | – |
| o4 Mini HighOpenAI · typical host price | $1.10 | $4.40 | $1.93 | one host | – | – | – |
| DeepSeek V4 Pro 0813DeepSeek · open weights · typical host price | $1.31 | $3.96 | $1.97 | $0.99Ionstream · 22 hosts | – | – | – |
| DeepSeek V4 Pro 0423DeepSeek · open weights · developer priceHalf price off-peak ($0.66 / $1.98) | $1.32 | $3.96 | $1.98 | $0.26StreamLake, FP8 · 15 hosts | 36.0 | 31 | 115 |
| Claude Haiku 4.5Anthropic · developer price | $1.00 | $5.00 | $2.00 | $2.00Azure · 8 hosts | 16.9 | 8 | 102 |
| Muse Spark 1.1Meta · typical host price | $1.25 | $4.25 | $2.00 | one host | 33.7 | – | – |
| Muse Spark 1.2Meta · typical host price | $1.25 | $4.25 | $2.00 | one host | 39.6 | – | – |
| Muse Spark 1.3Meta · developer priceAlso $0.10 / $0.20 if Meta may train on your data | $1.25 | $4.25 | $2.00 | one host | 48.1 | 66 | 137 |
| GLM 5.1Z.ai · open weights · typical host price | $1.38 | $4.40 | $2.13 | $1.48StreamLake, FP8 · 13 hosts | 26.1 | – | – |
| GLM 5.2Z.ai · open weights · developer price | $1.40 | $4.40 | $2.15 | $0.87DeepInfra, FP4 · 33 hosts | 33.7 | – | – |
| GLM 5.3Z.ai · open weights · developer price | $1.40 | $4.40 | $2.15 | $0.65Novita, FP8 · 41 hosts | 44.8 | 39 | 75 |
| Qwen3.7 MaxAlibaba · typical host price | $1.48 | $4.42 | $2.21 | one host | 29.5 | – | – |
| Grok 4.5xAI · developer price | $2.00 | $6.00 | $3.00 | $3.00xAI · 4 hosts | 38.8 | – | – |
| Grok 4.6xAI · developer price | $2.00 | $6.00 | $3.00 | $3.00xAI · 6 hosts | 44.3 | – | – |
| Grok 4.7xAI · developer price | $2.00 | $6.00 | $3.00 | $3.00xAI · 5 hosts | 46.4 | 43 | 68 |
| Mistral Large 2407Mistral · typical host price | $2.00 | $6.00 | $3.00 | $3.00Mistral · 3 hosts | 6.8 | – | – |
| Mistral Medium 3.5Mistral · developer price | $1.50 | $7.50 | $3.00 | $3.00Mistral · 3 hosts | 14.2 | 21 | 163 |
| Mixtral 8x22B InstructMistral · open weights · typical host price | $2.00 | $6.00 | $3.00 | $3.00Mistral · 3 hosts | 5.7 | – | – |
| Qwen3.8 2.4T A95BAlibaba · open weights · developer price | $2.00 | $6.00 | $3.00 | $3.00Novita · 7 hosts | 39.9 | – | 38 |
| Qwen3.8 Max (0902)Alibaba · developer price | $2.00 | $6.00 | $3.00 | one host | 45.4 | 46 | 39 |
| Gemini 3.5 FlashGoogle · developer price | $1.50 | $9.00 | $3.38 | $1.69Google · 7 hosts | 33.6 | – | – |
| GPT-5OpenAI · typical host price | $1.25 | $10.00 | $3.44 | $3.44Azure · 3 hosts | 23.0 | – | – |
| GPT-5.1OpenAI · typical host price | $1.25 | $10.00 | $3.44 | $1.72OpenAI · 5 hosts | 24.7 | – | – |
| GPT-5.1-CodexOpenAI · typical host price | $1.25 | $10.00 | $3.44 | one host | 23.7 | – | – |
| Gemini 2.5 Pro Preview 06-05Google · typical host price | $1.25 | $10.00 | $3.44 | $1.72Google AI Studio · 7 hosts | – | – | – |
| GPT-4.1OpenAI · typical host price | $2.00 | $8.00 | $3.50 | $3.50Azure · 3 hosts | 12.7 | – | – |
| o3OpenAI · typical host price | $2.00 | $8.00 | $3.50 | one host | 20.2 | – | 126 |
| Claude Sonnet 5Anthropic · developer price | $2.00 | $10.00 | $4.00 | $4.00Claude Platform on AWS · 10 hosts | 38.2 | – | – |
| Claude Sonnet 5.5Anthropic · developer price | $2.00 | $10.00 | $4.00 | $4.00Google · 8 hosts | 56.0 | 64 | 137 |
| GPT-6 SolOpenAI · developer price | $2.00 | $10.00 | $4.00 | $2.00OpenAI · 7 hosts | 47.6 | 69 | – |
| GPT-6.1 SolOpenAI · developer price | $2.00 | $10.00 | $4.00 | $2.00OpenAI · 7 hosts | 51.8 | 88 | 57 |
| Command ACohere · open weights · typical host price | $2.50 | $10.00 | $4.38 | one host | 7.0 | – | 61 |
| Command R+ (08-2024)Cohere · typical host price | $2.50 | $10.00 | $4.38 | one host | – | – | – |
| GPT-4o (2024-08-06)OpenAI · typical host price | $2.50 | $10.00 | $4.38 | $4.38Azure · 2 hosts | 7.7 | – | – |
| GPT-5.6 TerraOpenAI · developer price | $2.00 | $12.00 | $4.50 | $2.25OpenAI · 7 hosts | 42.1 | – | 104 |
| Gemini 3.1 Pro PreviewGoogle · developer price | $2.00 | $12.00 | $4.50 | $2.25Google · 6 hosts | 29.7 | 47 | 121 |
| Gemini 3.1 Pro Preview Custom ToolsGoogle · typical host price | $2.00 | $12.00 | $4.50 | one host | – | – | – |
| GPT-5.2OpenAI · typical host price | $1.75 | $14.00 | $4.81 | $2.41OpenAI · 4 hosts | 30.4 | – | – |
| GPT-5.2-CodexOpenAI · typical host price | $1.75 | $14.00 | $4.81 | one host | 28.5 | – | – |
| GPT-5.3-CodexOpenAI · typical host price | $1.75 | $14.00 | $4.81 | $4.81Azure · 3 hosts | 32.5 | – | 94 |
| GPT-5.4OpenAI · developer price | $2.50 | $15.00 | $5.62 | $2.81OpenAI · 7 hosts | 39.0 | – | – |
| Claude Sonnet 4Anthropic · typical host price | $3.00 | $15.00 | $6.00 | $6.00Amazon Bedrock · 2 hosts | 18.9 | – | – |
| Claude Sonnet 4.5Anthropic · typical host price | $3.00 | $15.00 | $6.00 | $6.00Claude Platform on AWS · 7 hosts | 20.7 | – | – |
| Claude Sonnet 4.6Anthropic · typical host price | $3.00 | $15.00 | $6.00 | $6.00Claude Platform on AWS · 9 hosts | 30.1 | – | – |
| Kimi K3Moonshot AI · open weights · developer price | $3.00 | $15.00 | $6.00 | $3.71Relace, FP4 · 24 hosts | 43.6 | 49 | 45 |
| GPT-4o (2024-05-13)OpenAI · typical host price | $5.00 | $15.00 | $7.50 | $7.50Azure · 2 hosts | 7.3 | – | – |
| Claude Opus 5.5Anthropic · developer price | $4.00 | $20.00 | $8.00 | $8.00Amazon Bedrock · 11 hosts | 57.6 | 82 | 97 |
| GPT-5.6 SolOpenAI · developer pricePromotional price until at least 21 Nov 2026 | $4.00 | $20.00 | $8.00 | $2.00OpenAI · 7 hosts | 47.0 | – | – |
| GPT-5.6 Sol ProOpenAI · typical host price | $4.00 | $20.00 | $8.00 | $2.00OpenAI · 5 hosts | – | – | – |
| Claude Opus 4.5Anthropic · typical host price | $5.00 | $25.00 | $10.00 | $10.00Claude Platform on AWS · 6 hosts | 29.1 | – | – |
| Claude Opus 4.6Anthropic · typical host price | $5.00 | $25.00 | $10.00 | $10.00Claude Platform on AWS · 6 hosts | 31.9 | – | – |
| Claude Opus 4.7Anthropic · typical host price | $5.00 | $25.00 | $10.00 | $10.00Claude Platform on AWS · 8 hosts | 40.7 | – | – |
| Claude Opus 5Anthropic · developer price | $5.00 | $25.00 | $10.00 | $10.00Claude Platform on AWS · 11 hosts | 50.8 | – | – |
| Claude Opus 4.8Anthropic · typical host price | $5.50 | $27.50 | $11.00 | $10.00Claude Platform on AWS · 11 hosts | 41.8 | – | – |
| GPT-5.5OpenAI · developer price | $5.00 | $30.00 | $11.25 | $5.62OpenAI · 7 hosts | 38.4 | – | – |
| GPT-4 TurboOpenAI · typical host price | $10.00 | $30.00 | $15.00 | one host | – | – | – |
| Claude Fable 5Anthropic · developer price | $10.00 | $50.00 | $20.00 | $20.00Claude Platform on AWS · 6 hosts | 49.6 | – | – |
| Claude Fable 5.1Anthropic · developer price | $10.00 | $50.00 | $20.00 | $20.00Azure · 4 hosts | 53.4 | 65 | 71 |
| GPT-6 AstraOpenAI · developer price | $10.00 | $50.00 | $20.00 | $10.00OpenAI · 7 hosts | 52.7 | 78 | 52 |
| o1OpenAI · typical host price | $15.00 | $60.00 | $26.25 | one host | 15.2 | – | – |
| Claude Opus 4.1Anthropic · typical host price | $15.00 | $75.00 | $30.00 | one host | 22.8 | – | – |
| o3 ProOpenAI · typical host price | $20.00 | $80.00 | $35.00 | one host | 21.9 | – | – |
| GPT-5.4 ProOpenAI · typical host price | $30.00 | $180.00 | $67.50 | $33.75OpenAI · 3 hosts | – | – | – |
The cheapest model at each level
Grouped by the Artificial Analysis Intelligence Index. The cheapest model in a group is rarely the most capable one, but it is often close.
| Intelligence | Cheapest | Price | Most capable | Price |
|---|---|---|---|---|
| Frontier: 50 or more | Claude Sonnet 5.5Anthropic · 56.0 | $4.00 | Claude Opus 5.5Anthropic · 57.6 | $8.00 |
| Strong: 40–50 | GLM 5.3 FlashZ.ai · 41.8 | $0.24 | Claude Fable 5Anthropic · 49.6 | $20.00 |
| Capable: 30–40 | DeepSeek V4 Flash 0423DeepSeek · 34.3 | $0.18 | Qwen3.8 2.4T A95BAlibaba · 39.9 | $3.00 |
| Everyday: 20–30 | Solar Pro 4Upstage · 28.2 | $0.16 | Gemini 3.1 Pro PreviewGoogle · 29.7 | $4.50 |
| Light: under 20 | Llama 3.1 8B InstructMeta · 6.9 | $0.058 | Qwen3.5-35B-A3BAlibaba · 19.3 | $0.45 |
Price against intelligence
What each model costs for what it can do
Further up is more capable; further left is cheaper. The labelled models are the best value: none is beaten on both.
Each dot is a model on sale today. Labelled dots are the best-value models: for each, no other model is both cheaper and more capable. Price is on a log scale.
Best and worst value
Best value
Each one is the cheapest way to reach its level of intelligence.
- Claude Opus 5.5 57.6 for $8.00
- Claude Sonnet 5.5 56.0 for $4.00
- Muse Spark 1.3 48.1 for $2.00
- MiMo-V2.6-Pro 46.3 for $0.54
- GLM 5.3 Flash 41.8 for $0.24
- GPT-6 Luna 38.1 for $0.20
- MiMo-V2.6-Flash 37.9 for $0.18
- DeepSeek V4 Flash 0423 34.3 for $0.18
- Solar Pro 4 28.2 for $0.16
- Gemma 4 26B A4B 16.7 for $0.15
- GLM 4.7 Flash 14.9 for $0.15
- Qwen3.5-9B 13.3 for $0.11
- Granite 4.2 8B 11.1 for $0.11
- gpt-oss-20b 10.0 for $0.060
- Llama 3.1 8B Instruct 6.9 for $0.058
Expensive for what you get
At least 3× the price of the cheapest model that is as capable, on the Intelligence Index alone. A model can still be worth it for a task the index doesn't measure: check its model page.
- o3 Pro $35.00: 222× Solar Pro 4 ($0.16, 28.2)
- Claude Opus 4.1 $30.00: 190× Solar Pro 4 ($0.16, 28.2)
- o1 $26.25: 175× Gemma 4 26B A4B ($0.15, 16.7)
- GPT-4o (2024-05-13) $7.50: 125× gpt-oss-20b ($0.060, 10.0)
- Command A $4.38: 73× gpt-oss-20b ($0.060, 10.0)
- GPT-4o (2024-08-06) $4.38: 73× gpt-oss-20b ($0.060, 10.0)
- Claude Opus 4.5 $10.00: 57× DeepSeek V4 Flash 0423 ($0.18, 34.3)
- Claude Opus 4.6 $10.00: 57× DeepSeek V4 Flash 0423 ($0.18, 34.3)
- Mistral Large 2407 $3.00: 52× Llama 3.1 8B Instruct ($0.058, 6.9)
- Mixtral 8x22B Instruct $3.00: 52× Llama 3.1 8B Instruct ($0.058, 6.9)
- GPT-5.5 $11.25: 47× GLM 5.3 Flash ($0.24, 41.8)
- Claude Opus 4.8 $11.00: 46× GLM 5.3 Flash ($0.24, 41.8)
Price per token isn't cost per task
A model that writes three times as many tokens costs three times as much, whatever its price list says. These are the costs we measured running CatalogBench: turning a sparse product feed into listings, at each model's cheapest setting.
| Model | Blended price | Rank by price | Measured cost per 10,000 listings | Rank by cost |
|---|---|---|---|---|
| GPT-6 Luna | $0.20 | 1 | $4.77 | 1 |
| Gemini 3.5 Flash Lite | $0.85 | 3 | $20.47 | 2 |
| Mistral Large 4 | $1.03 | 4 | $32.77 | 3 |
| DeepSeek V4.1 Flash | $0.52 | 2 | $60.70 | 4 |
| Claude Haiku 4.5 | $2.00 | 7 | $61.38 | 5 |
| GLM 5V Turbo | $1.90 | 6 | $70.11 | 6 |
| Mistral Medium 3.5 | $3.00 | 10 | $80.05 | 7 ▲ cheaper than its price suggests |
| GPT-6.1 Sol | $4.00 | 14 | $83.84 | 8 ▲ cheaper than its price suggests |
| GPT-6 Sol | $4.00 | 13 | $85.18 | 9 ▲ cheaper than its price suggests |
| Gemini 3.8 Flash | $1.50 | 5 | $102.92 | 10 ▼ dearer than its price suggests |
| Muse Spark 1.3 | $2.00 | 8 | $164.18 | 11 ▼ dearer than its price suggests |
| Claude Sonnet 5.5 | $4.00 | 12 | $171.91 | 12 |
| Grok 4.7 | $3.00 | 9 | $237.19 | 13 ▼ dearer than its price suggests |
| Qwen3.8 Max (0902) | $3.00 | 11 | $255.72 | 14 ▼ dearer than its price suggests |
| Gemini 3.1 Pro Preview | $4.50 | 15 | $354.64 | 15 |
| GPT-6 Astra | $20.00 | 19 | $371.20 | 16 ▲ cheaper than its price suggests |
| Claude Opus 5.5 | $8.00 | 17 | $400.06 | 17 |
| Kimi K3 | $6.00 | 16 | $468.95 | 18 |
| Claude Fable 5.1 | $20.00 | 18 | $991.28 | 19 |
How to cut your LLM bill
Choose by cost per task, not price per token
Verbose models cost more than their price suggests. Run your own task on two or three candidates and compare what each run cost, as in the table above.
Use the smallest model that passes
Define what "good enough" means for your task, then test downwards. The cheapest model in each intelligence band is often within a few points of the best.
Set the reasoning effort deliberately
High reasoning can multiply the tokens a model writes. On our benchmarks it sometimes helps and sometimes makes results worse: Mistral Large 4 cost almost four times as much per listing at high reasoning, with fewer invented claims but more failures.
Cache the prompt you repeat
Most developers charge far less for cached input. Put instructions, examples and reference data at the start of the prompt, unchanged, so they can be cached.
Batch what can wait
Batch APIs from OpenAI, Anthropic, Google and others take about half off for work returned within a day: catalogue enrichment, evaluations, backfills.
Route by difficulty
Send easy requests to a cheap model and escalate only what it can't handle. A simple check on the cheap model's output often decides.
Cap the output
Ask for structured output and set a maximum length. Output tokens typically cost three to six times as much as input.
Watch launch discounts end
New models often launch at half price for a few weeks. Budget at the list price.
Quick answers
What is the cheapest LLM API?
Among models with an Artificial Analysis Intelligence Index score, Llama 3.1 8B Instruct is the cheapest, at $0.058 per million tokens (three input tokens to one output). Cheaper models exist, but they are much less capable.
What is the cheapest capable LLM?
GLM 5.3 Flash is the cheapest model scoring 40 or more on the Intelligence Index, at $0.24 per million tokens.
How much does the most capable model cost?
Claude Opus 5.5 scores highest on the Intelligence Index (57.6) and costs $4.00 per million input tokens and $20.00 per million output tokens.
Why do output tokens cost more than input tokens?
Generating each output token takes a full pass through the model, while input tokens are processed together. Across the models we track, output costs a median of four times as much as input.
Is self-hosting an open model cheaper than an API?
Usually not, unless you run a high, steady volume. A server big enough for a strong open model costs thousands of dollars a month on AWS, while hosted APIs for the same models are often well under a dollar per million tokens. The table above shows the break-even volume for each model.
What does 'blended price' mean?
One price per million tokens for a typical mix of three input tokens to one output, so models can be compared on a single number. Your own mix may differ: chat and summarisation are input-heavy, generation is output-heavy.
Which model is cheapest for your task?
Run the models you're considering side by side on your own prompt, with the cost of every answer. Or ask us to find the cheapest model that's good enough on your real data.
Sources and method
- Developer prices: read by hand from each developer's own pricing page, checked 7 Oct 2026; each links from its row in the table. Launch discounts and their end dates are noted.
- Host prices: every host's price for each model from OpenRouter's public endpoints list, checked 7 Oct 2026. "Typical host" is the median; "cheapest host" is the lowest blended price, with the model's quantisation where the host states it.
- Intelligence and speed: Artificial Analysis (data sourced from Artificial Analysis). Our overall: the Spring Prompt overall score.
- Self-hosting: model sizes from the developers' model cards, server prices from AWS's public on-demand price list (us-east-1), throughput only where a developer or serving project has published it.
- Measured cost per task: what our own CatalogBench runs cost, from provider billing, at each model's cheapest setting.