AI model price report — week 37, 2026
No price movements detected. This report is generated from the same dataset that powers the JSON API and the raw changelog — 1,809 priced chat models across 33 providers. Dataset refreshed 2026-09-13 13:40 UTC.
Price movements this week
No price change was detected in any refresh run during this week. The dataset is diffed against the previous snapshot on every run, so a quiet week means the providers we track held their published rates. Systematic tracking of price history began on 2026-09-13, so earlier weeks carry no comparison baseline.
1 refresh run this week · 0 total diffs recorded.
Models added
No new endpoints appeared in this week's refresh runs.
Models retired
No tracked endpoints disappeared from the upstream sources this week.
Cheapest input tokens
The floor of the market as of this data refresh — what the input side of a bill costs per 1M tokens.
| # | Model | Provider | Input $/1M | Output $/1M | Cached $/1M | Context |
|---|---|---|---|---|---|---|
| 1 | Qwen2.5 Coder 7B | Nebius | $0.01 | $0.03 | — | 33K |
| 2 | Qwen2.5 Coder 3B Instruct | Nscale | $0.01 | $0.03 | — | — |
| 3 | Qwen2.5 Coder 7B Instruct | Nscale | $0.01 | $0.03 | — | — |
| 4 | Nemotron 3.5 Lightning 30b A3b | Perplexity | $0.011 | $0.17 | $0.0011 | — |
| 5 | Granite 4.0 H Micro | Cloudflare Workers AI | $0.017 | $0.112 | — | 131K |
| 6 | Mistral Nemo Instruct 2407 | DeepInfra | $0.019 | $0.03 | — | 131K |
| 7 | Mistral Nemo | OpenRouter | $0.019 | $0.03 | — | 131K |
| 8 | Llama 3.2 3B Instruct | DeepInfra | $0.02 | $0.02 | — | 131K |
| 9 | Meta Llama 3.1 8B Instruct Turbo | DeepInfra | $0.02 | $0.04 | — | 131K |
| 10 | Gemma 4 E4B It | DeepInfra | $0.02 | $0.1 | — | 131K |
| 11 | Llama Guard 3 8B | Nebius | $0.02 | $0.06 | — | 128K |
| 12 | Meta Llama 3.1 8B Instruct | Nebius | $0.02 | $0.06 | — | 128K |
Best blended cost
Weighted 3:1 toward input tokens, which is how most chat and agent workloads actually bill.
| # | Model | Provider | Input $/1M | Output $/1M | Cached $/1M | Context |
|---|---|---|---|---|---|---|
| 1 | Gemini Exp 1114 | Google Gemini | free | free | — | 1.0M |
| 2 | Gemini Exp 1206 | Google Gemini | free | free | — | 2.1M |
| 3 | Gemma 3 27b It | Google Gemini | free | free | — | 131K |
| 4 | Gemma 4 26b A4b It | Google Gemini | free | free | — | 262K |
| 5 | Gemma 4 31b It | Google Gemini | free | free | — | 262K |
| 6 | Learnlm 1.5 Pro Experimental | Google Gemini | free | free | — | 33K |
| 7 | Labs Leanstral 1 5 | Mistral AI | free | free | — | 262K |
| 8 | Labs Leanstral 1 5 1 | Mistral AI | free | free | — | 262K |
| 9 | Anthropic.claude Mythos Preview | Amazon Bedrock | free | free | — | 1M |
| 10 | Gemma 2b It Lora | Cloudflare Workers AI | free | free | — | 8K |
| 11 | Mistral 7b Instruct V0.2 Lora | Cloudflare Workers AI | free | free | — | 15K |
| 12 | Llama 2 7b Chat Hf Lora | Cloudflare Workers AI | free | free | — | 8K |
Most expensive models tracked
The top of the market, for reference — what frontier pricing looks like against the floor above.
| # | Model | Provider | Input $/1M | Output $/1M | Cached $/1M | Context |
|---|---|---|---|---|---|---|
| 1 | O1 Pro | OpenRouter | $150 | $600 | — | 200K |
| 2 | O1 Pro | OpenAI | $150 | $600 | — | 200K |
| 3 | O1 Pro | OpenAI | $150 | $600 | — | 200K |
| 4 | GPT 4.5 Preview | Azure OpenAI | $75 | $150 | $37.5 | 128K |
| 5 | GPT 4 32k 0613 | Azure OpenAI | $60 | $120 | — | 33K |
| 6 | GPT 4 32k | Azure OpenAI | $60 | $120 | — | 33K |
| 7 | GPT 5.4 Pro | OpenRouter | $30 | $180 | — | 1.1M |
| 8 | GPT 5.5 Pro | OpenRouter | $30 | $180 | — | 1.1M |
What this costs in practice
Three common workloads priced at the cheapest rate available in the index this week.
| Workload | Priced at | Total |
|---|---|---|
| support bot · 10,000 chats/day for a month (1.2k in, 400 out) | Qwen2.5 Coder 7B · $0.01/$0.03 | $7.20 |
| 1,000,000 document summaries (8k in, 300 out) | Qwen2.5 Coder 7B · $0.01/$0.03 | $89.00 |
| RAG index · 100M embedding tokens | Pplx Embed V1 0.6b · $0.004/free | $0.40 |
Straight arithmetic from published list prices — no volume discounts, batch tiers or committed-use pricing applied. Run your own numbers in the cost calculator.
Provider price floors
Median and floor input price per provider across the models we track, and how many moved this week.
| Provider | Models | Floor in | Median in | Top in | Moved this week |
|---|---|---|---|---|---|
| Fireworks AI | 272 | $0.05 | $0.22 | $4.5 | — |
| OpenRouter | 268 | $0.019 | $0.45 | $150 | — |
| DeepInfra | 135 | $0.019 | $0.27 | $16.5 | — |
| Novita AI | 132 | $0.02 | $0.28 | $4 | — |
| OpenAI | 112 | $0.05 | $2 | $150 | — |
| Amazon Bedrock | 109 | $0.042 | $0.99 | $18.8 | — |
| Azure OpenAI | 104 | $0.05 | $1.88 | $75 | — |
| Together AI | 78 | $0.05 | $0.5 | $3.5 | — |
| Mistral AI | 73 | $0.06 | $0.4 | $4 | — |
| Databricks | 67 | $0.05 | $1.36 | $30 | — |
| Perplexity | 66 | $0.011 | $1.25 | $10 | — |
| Nebius | 56 | $0.01 | $0.15 | $3 | — |
| Google Gemini | 49 | $0.075 | $0.35 | $2 | — |
| xAI | 44 | $1 | $1.25 | $2 | — |
| Replicate | 40 | $0.03 | $0.65 | $15 | — |
| Cloudflare Workers AI | 30 | $0.017 | $0.351 | $1.92 | — |
| Ollama | 29 | — | — | — | — |
| Anthropic | 28 | $0.25 | $5 | $15 | — |
| Alibaba DashScope | 25 | $0.05 | $0.4 | $2.5 | — |
| Moonshot AI | 24 | $0.2 | $1 | $3 | — |
Frequently asked
› What does the 2026-w37 report measure?
It summarises every price difference detected between consecutive refresh runs of the dataset. Each refresh snapshots input, output and cached-input rates per 1M tokens for every tracked endpoint, then diffs that snapshot against the previous one; the movements listed on this page are those diffs, grouped by ISO week.
› Why does a week sometimes show no changes?
Because providers do not change published rates every week. A report with zero movements means every diff run in that week came back clean — the figures in the market snapshot sections still reflect the current dataset.
› Can I reuse this report?
Yes. The underlying data is published under CC BY 4.0 and served free at /api/models. Attribute Price per 1M with a link when you republish figures.