LLM API Pricing Index
Input and output price per million tokens for 440 models from 59 vendors, in one schema. 418 are paid, 22 have a free tier. Blended price assumes 3 input tokens per output token, the ratio most chat and agent work lands near.
Last collected 2026-09-12 · 7 snapshots
The spread is the story: the frontier tier costs 13× the cheapest flagship-class model on this chart, for work that often does not need it.
Every model
| Context | Cache read | 7D | ||||
|---|---|---|---|---|---|---|
 Qwen3.7 FlashAlibaba Qwen · qwen/qwen3.7-flash | 1M | $0.030 | $0.130 | $0.055 | $0.0060 | — |
| Qwen3.5-9BAlibaba Qwen · qwen/qwen3.5-9b | 262K | $0.100 | $0.150 | $0.113 | — | — |
| Qwen3.5-FlashAlibaba Qwen · qwen/qwen3.5-flash-02-23 | 1M | $0.065 | $0.260 | $0.114 | — | — |
| Qwen3 Coder 30B A3B InstructAlibaba Qwen · qwen/qwen3-coder-30b-a3b-instruct | 262K | $0.070 | $0.280 | $0.122 | — | — |
| Qwen2.5 7B InstructAlibaba Qwen · qwen/qwen-2.5-7b-instruct | 33K | $0.100 | $0.200 | $0.125 | — | — |
| Qwen3 32BAlibaba Qwen · qwen/qwen3-32b | 131K | $0.080 | $0.280 | $0.130 | — | — |
| Qwen3 30B A3B Instruct 2507Alibaba Qwen · qwen/qwen3-30b-a3b-instruct-2507 | 262K | $0.090 | $0.300 | $0.142 | — | — |
| Qwen3 235B A22B Instruct 2507Alibaba Qwen · qwen/qwen3-235b-a22b-2507 | 262K | $0.087 | $0.350 | $0.153 | $0.018 | — |
| Qwen3 VL 32B InstructAlibaba Qwen · qwen/qwen3-vl-32b-instruct | 131K | $0.104 | $0.416 | $0.182 | — | — |
| Qwen3.5-9B (batch)Alibaba Qwen · qwen/qwen3.5-9b:batch | 262K | $0.170 | $0.250 | $0.190 | — | — |
| Qwen3 VL 8B InstructAlibaba Qwen · qwen/qwen3-vl-8b-instruct | 262K | $0.117 | $0.455 | $0.202 | — | — |
| Qwen3 8BAlibaba Qwen · qwen/qwen3-8b | 131K | $0.117 | $0.455 | $0.202 | — | — |
| Qwen3 30B A3BAlibaba Qwen · qwen/qwen3-30b-a3b | 131K | $0.120 | $0.500 | $0.215 | — | — |
| Qwen3.8 FlashAlibaba Qwen · qwen/qwen3.8-flash | 1M | $0.150 | $0.470 | $0.230 | $0.016 | — |
| Qwen3 VL 30B A3B InstructAlibaba Qwen · qwen/qwen3-vl-30b-a3b-instruct | 262K | $0.150 | $0.600 | $0.263 | — | — |
| Qwen3 Coder NextAlibaba Qwen · qwen/qwen3-coder-next | 262K | $0.120 | $0.800 | $0.290 | $0.070 | — |
| Qwen3.6 35B A3BAlibaba Qwen · qwen/qwen3.6-35b-a3b | 262K | $0.100 | $0.900 | $0.300 | $0.050 | — |
| Qwen3 Next 80B A3B InstructAlibaba Qwen · qwen/qwen3-next-80b-a3b-instruct | 262K | $0.090 | $1.10 | $0.343 | — | — |
| Qwen2.5 72B InstructAlibaba Qwen · qwen/qwen-2.5-72b-instruct | 33K | $0.360 | $0.400 | $0.370 | — | — |
| Qwen3 Coder FlashAlibaba Qwen · qwen/qwen3-coder-flash | 1M | $0.195 | $0.975 | $0.390 | $0.039 | — |
| Qwen Plus 0728Alibaba Qwen · qwen/qwen-plus-2025-07-28 | 1M | $0.260 | $0.780 | $0.390 | — | — |
| Qwen-PlusAlibaba Qwen · qwen/qwen-plus | 1M | $0.260 | $0.780 | $0.390 | $0.052 | — |
| Qwen3 14BAlibaba Qwen · qwen/qwen3-14b | 131K | $0.228 | $0.910 | $0.398 | — | — |
| Qwen3 Next 80B A3B ThinkingAlibaba Qwen · qwen/qwen3-next-80b-a3b-thinking | 262K | $0.150 | $1.20 | $0.412 | — | — |
| Qwen3.6 FlashAlibaba Qwen · qwen/qwen3.6-flash | 1M | $0.188 | $1.13 | $0.422 | — | — |
| Qwen3 Coder 480B A35BAlibaba Qwen · qwen/qwen3-coder | 262K | $0.300 | $1.00 | $0.475 | $0.100 | — |
| Qwen3.5-27BAlibaba Qwen · qwen/qwen3.5-27b | 262K | $0.195 | $1.56 | $0.536 | — | — |
| Qwen3.5-35B-A3BAlibaba Qwen · qwen/qwen3.5-35b-a3b | 262K | $0.313 | $1.25 | $0.547 | $0.156 | — |
| Qwen3.7 PlusAlibaba Qwen · qwen/qwen3.7-plus | 1M | $0.320 | $1.28 | $0.560 | $0.064 | — |
| Qwen3.5 Plus 2026-02-15Alibaba Qwen · qwen/qwen3.5-plus-02-15 | 1M | $0.260 | $1.56 | $0.585 | — | — |
| Qwen3 VL 235B A22B InstructAlibaba Qwen · qwen/qwen3-vl-235b-a22b-instruct | 262K | $0.210 | $1.90 | $0.632 | $0.100 | — |
| Qwen3 VL 8B ThinkingAlibaba Qwen · qwen/qwen3-vl-8b-thinking | 131K | $0.180 | $2.10 | $0.660 | — | — |
| Qwen3.5 Plus 2026-04-20Alibaba Qwen · qwen/qwen3.5-plus-20260420 | 1M | $0.300 | $1.80 | $0.675 | — | — |
| Qwen3.5-122B-A10BAlibaba Qwen · qwen/qwen3.5-122b-a10b | 262K | $0.260 | $2.08 | $0.715 | — | — |
| Qwen3.6 27BAlibaba Qwen · qwen/qwen3.6-27b | 262K | $0.300 | $2.00 | $0.725 | $0.030 | — |
| Qwen3.6 PlusAlibaba Qwen · qwen/qwen3.6-plus | 1M | $0.325 | $1.95 | $0.731 | — | — |
| Qwen2.5 Coder 32B InstructAlibaba Qwen · qwen/qwen-2.5-coder-32b-instruct | 33K | $0.660 | $1.00 | $0.745 | — | — |
| Qwen3 235B A22B Thinking 2507Alibaba Qwen · qwen/qwen3-235b-a22b-thinking-2507 | 131K | $0.230 | $2.30 | $0.748 | — | — |
| Qwen3 VL 30B A3B ThinkingAlibaba Qwen · qwen/qwen3-vl-30b-a3b-thinking | 262K | $0.200 | $2.40 | $0.750 | — | — |
| Qwen3 30B A3B Thinking 2507Alibaba Qwen · qwen/qwen3-30b-a3b-thinking-2507 | 82K | $0.200 | $2.40 | $0.750 | — | — |
| Qwen3 235B A22BAlibaba Qwen · qwen/qwen3-235b-a22b | 131K | $0.455 | $1.82 | $0.796 | — | — |
| Qwen3.8 27BAlibaba Qwen · qwen/qwen3.8-27b | 1M | $0.214 | $2.55 | $0.798 | $0.150 | — |
| Qwen2.5 VL 72B InstructAlibaba Qwen · qwen/qwen2.5-vl-72b-instruct | 128K | $0.800 | $1.00 | $0.850 | $0.400 | — |
| Qwen3.5 397B A17BAlibaba Qwen · qwen/qwen3.5-397b-a17b | 262K | $0.550 | $3.50 | $1.29 | $0.225 | — |
| Qwen3 VL 235B A22B ThinkingAlibaba Qwen · qwen/qwen3-vl-235b-a22b-thinking | 131K | $0.400 | $4.00 | $1.30 | — | — |
| Qwen3 Coder PlusAlibaba Qwen · qwen/qwen3-coder-plus | 1M | $0.650 | $3.25 | $1.30 | $0.130 | — |
| Qwen3 Max ThinkingAlibaba Qwen · qwen/qwen3-max-thinking | 262K | $0.780 | $3.90 | $1.56 | — | — |
| Qwen3 MaxAlibaba Qwen · qwen/qwen3-max | 262K | $0.780 | $3.90 | $1.56 | $0.156 | — |
| Qwen3.7 MaxAlibaba Qwen · qwen/qwen3.7-max | 1M | $1.48 | $4.42 | $2.21 | $0.295 | — |
| Qwen3.6 Max PreviewAlibaba Qwen · qwen/qwen3.6-max-preview | 262K | $1.03 | $6.16 | $2.31 | — | — |
| Qwen3.8 Max (0902)Alibaba Qwen · qwen/qwen3.8-max-0902 | 1M | $2.00 | $6.00 | $3.00 | $0.250 | — |
| Qwen3.8 2.4T A95BAlibaba Qwen · qwen/qwen3.8-2.4t-a95b | 1.0M | $2.00 | $6.00 | $3.00 | $0.250 | — |
| Qwen3.8 2.4T A95B (batch)Alibaba Qwen · qwen/qwen3.8-2.4t-a95b:batch | 1.0M | $2.00 | $6.00 | $3.00 | $0.250 | — |
53 of 53 models (free tiers hidden). Blended = 3 input tokens per output token. Prices as listed on OpenRouter on the date above.
Vendors
| Vendor | Models | Cheapest paid, blended |
|---|---|---|
 OpenAI | 93 | $0.055 |
| Alibaba Qwen | 53 | $0.055 |
 Google | 43 | $0.063 |
 Anthropic | 27 | $0.500 |
 Mistral | 25 | $0.022 |
| 18 | $0.050 | |
 Z.ai (Zhipu) | 17 | $0.119 |
| 10 | $0.087 | |
| 9 | $0.425 | |
| 8 | $0.900 | |
 Meta | 8 | $0.058 |
| MMeta | 7 | $0.125 |
 Tencent | 7 | $0.077 |
 xAI | 7 | $1.25 |
| iinclusionAI | 6 | $0.032 |
| BBytedance Seed | 6 | $0.131 |
 Thinking Machines | 6 | $0.637 |
| ~~Openai | 5 | $0.450 |
 Cohere | 5 | $0.066 |
 Amazon | 5 | $0.061 |
 Perplexity | 5 | $1.00 |
 Sakana AI | 4 | $1.71 |
| PPoolside | 4 | $0.075 |
| AAion Labs | 4 | $0.875 |
| ~~Anthropic | 4 | $2.00 |
| TThedrummer | 3 | $0.350 |
 Nous Research | 3 | $0.700 |
| SSao10k | 3 | $0.043 |
| IInference Net | 2 | $0.060 |
| IInception | 2 | $0.068 |
| NNex Agi | 2 | — |
| IIbm Granite | 2 | $0.041 |
| ~~Z Ai | 2 | $0.119 |
| UUpstage | 2 | $0.158 |
| KKwaipilot | 2 | $0.525 |
| SStepfun | 2 | $0.150 |
| 2 | $1.50 | |
| XXiaomi | 2 | $0.175 |
| RRekaai | 2 | $0.100 |
| RRelace | 2 | $0.950 |
| MMorph | 2 | $0.900 |
| 2 | $0.087 | |
| DDots Studio | 1 | — |
 Liquid AI | 1 | — |
| ~~Deepseek | 1 | $0.040 |
| MMeituan | 1 | $0.525 |
| ~~X Ai | 1 | $3.00 |
| PPerceptron | 1 | $0.487 |
| ~~Moonshotai | 1 | $5.11 |
| AArcee Ai | 1 | $0.388 |
| OOpenrouter | 1 | — |
| WWriter | 1 | $1.95 |
| BBytedance | 1 | $0.125 |
| CCognitivecomputations | 1 | $0.375 |
| 1 | $0.627 | |
| AAnthracite Org | 1 | $3.13 |
| MMancer | 1 | $0.487 |
| UUndi95 | 1 | $0.425 |
| GGryphe | 1 | $0.060 |
How this is built
Prices are read from OpenRouter's public model list, which routes to each vendor and publishes the per-token price it charges — for nearly every model, the vendor's own list price passed through. It is the one place all vendors appear in one schema, which is why it is the source. Prices are shown as listed there; a vendor's direct price can differ, especially with volume discounts or provisioned throughput, and OpenRouter's own listing is the reference for what is shown.
Every collector run is stored as a snapshot and committed to the repository, so price changes are diffs anyone can inspect. Nothing is backfilled. The 7-day column compares a model against itself a week earlier and is blank until that run exists.
Blended weights input at 3 and output at 1. Change the ratio and the ranking changes — an output-heavy workload makes the frontier tier relatively more expensive, since output tokens cost five times input on most frontier models.