Track how AI model pricing changes week over week. Data from 3 weekly snapshots.
Price data sourced from OpenRouter API, captured in weekly snapshots. Currently tracking 3 weeks of data — trends become more meaningful as history grows.
Average and median pricing across all paid models for each tracked week. Prices are per 1M tokens.
| Week | Date Range | Avg Input $/1M | Avg Output $/1M | Median Input | Median Output | Free Models | Cheapest Paid |
|---|---|---|---|---|---|---|---|
| 2026-w08 | 2026-02-16 to 2026-02-22 | $2.22 | $9.26 | $0.30 | $1.20 | 23 | LFM2-8B-A1B(Liquid AI) |
| 2026-w09 | 2026-02-23 to 2026-03-01 | $2.21 | $9.24 | $0.30 | $1.20 | 21 | LFM2-8B-A1B(Liquid AI) |
| 2026-w10 | 2026-03-02 to 2026-03-08 | $2.20 | $9.21 | $0.30 | $1.20 | 25 | LFM2-8B-A1B(Liquid AI) |
Comparing 2026-w08 to 2026-w10 — models with the biggest pricing movements.
| Model | Provider | Input Before | Input After | Input Change | Output Before | Output After | Output Change |
|---|---|---|---|---|---|---|---|
| Command R (08-2024) | Cohere | $2.50 | $0.15 | -2.35 | $10.00 | $0.60 | -9.40 |
| Gemma 3 27B | $0.04 | $0.03 | -0.01 | $0.15 | $0.11 | -0.04 |
Per-provider pricing breakdown from the latest snapshot (2026-w10). Prices per 1M tokens.
| Provider | Models | Avg Input $/1M | Avg Output $/1M | Cheapest Model | Free Models |
|---|---|---|---|---|---|
| OpenAI | 63 | $7.26 | $30.18 | gpt-oss-20b | 2 |
| Alibaba | 51 | $0.25 | $1.12 | Qwen2.5 Coder 7B Instruct | 5 |
| 29 | $0.67 | $4.23 | Gemma 3n 4B | 5 | |
| Mistral AI | 25 | $0.60 | $1.83 | Mistral Nemo | 1 |
| Meta | 17 | $0.65 | $0.77 | Llama 3.1 8B Instruct | 2 |
| Anthropic | 13 | $4.85 | $24.25 | Claude 3 Haiku | 0 |
| DeepSeek | 12 | $0.35 | $0.98 | R1 Distill Qwen 32B | 0 |
| xAI | 8 | $1.27 | $6.06 | Grok 4.1 Fast | 0 |
| NVIDIA | 8 | $0.32 | $0.51 | Nemotron Nano 9B V2 | 3 |
| MiniMax | 7 | $0.29 | $1.28 | MiniMax M2.1 | 0 |
| arcee-ai | 7 | $0.47 | $1.13 | Trinity Mini | 2 |
| Allen AI | 7 | $0.14 | $0.34 | Olmo 2 32B Instruct | 0 |
| Moonshot AI | 5 | $0.49 | $2.18 | Kimi K2 0905 | 0 |
| Amazon | 5 | $0.74 | $3.72 | Nova Micro 1.0 | 0 |
| Perplexity | 5 | $2.20 | $9.40 | Sonar | 0 |
| Baidu | 5 | $0.20 | $0.69 | ERNIE 4.5 21B A3B Thinking | 0 |
| Liquid AI | 5 | $0.02 | $0.05 | LFM2-8B-A1B | 2 |
| ByteDance | 4 | $0.13 | $0.73 | UI-TARS 7B | 0 |
| Cohere | 4 | $1.30 | $5.19 | Command R7B (12-2024) | 0 |
| Inception | 3 | $0.25 | $0.75 | Mercury 2 | 0 |
| aion-labs | 3 | $1.83 | $3.67 | Aion-1.0-Mini | 0 |
| StepFun | 2 | $0.10 | $0.30 | Step 3.5 Flash | 1 |
| Kuaishou | 2 | $0.21 | $0.83 | KAT-Coder-Pro V1 | 0 |
| Microsoft | 2 | $0.34 | $0.38 | Phi 4 | 0 |
| Inflection | 2 | $2.50 | $10.00 | Inflection 3 Pi | 0 |
| Stability AI | 2 | — | — | — | 0 |
| Upstage | 1 | $0.15 | $0.60 | Solar Pro 3 | 0 |
| Xiaomi | 1 | $0.09 | $0.29 | MiMo-V2-Flash | 0 |
| essentialai | 1 | $0.15 | $0.15 | Rnj 1 Instruct | 0 |
| deepcogito | 1 | $1.25 | $1.25 | Cogito v2.1 671B | 0 |
| Writer | 1 | $0.60 | $6.00 | Palmyra X5 | 0 |
| Meituan | 1 | $0.20 | $0.80 | LongCat Flash Chat | 0 |
| AI21 Labs | 1 | $2.00 | $8.00 | Jamba Large 1.7 | 0 |
| Tencent | 1 | $0.14 | $0.57 | Hunyuan A13B Instruct | 0 |
| IBM | 1 | $0.02 | $0.11 | Granite 4.0 Micro | 0 |
| eleutherai | 1 | $0.80 | $1.20 | Llemma 7b | 0 |
| Midjourney | 1 | — | — | — | 0 |
| Black Forest Labs | 1 | — | — | — | 0 |
| Ideogram | 1 | — | — | — | 0 |
| Recraft | 1 | — | — | — | 0 |
| Adobe | 1 | — | — | — | 0 |
| Leonardo AI | 1 | — | — | — | 0 |
| Runway | 1 | — | — | — | 0 |
| Pika | 1 | — | — | — | 0 |
| Luma AI | 1 | — | — | — | 0 |
| Wan AI | 1 | — | — | — | 1 |
| Lightricks | 1 | — | — | — | 1 |
Pricing data is sourced from the OpenRouter API and captured in weekly snapshots. Each snapshot records input and output token costs per million for every available model, allowing us to track price changes over time across all providers.
Price drops typically occur when providers reduce costs due to infrastructure optimizations, increased competition, or strategic pricing decisions to gain market share. As AI hardware costs decrease and model serving becomes more efficient, many providers pass savings on to customers.
Average pricing sums all prices and divides by the number of models, which can be skewed by expensive outliers. Median pricing represents the middle value when all prices are sorted, giving a better sense of what a "typical" model costs. We show both to provide a complete picture of market pricing.