AI Rate CardCoding model API pricing: GPT-5.3 Codex, Kimi K2.7 Code, Qwen3 Coder and more
Updated Oct 25, 2026
Coding agents read a lot and write a lot, so the price per million tokens decides the bill quickly. Below are the models built for code that most providers sell, with the typical price (the middle of all providers) and the cheapest one. The tables are rebuilt from today's prices on every page view.
The models in short
Price per 1 million tokens, input / output, in the 24 October 2026 collection; "typical" is the median across providers:
- GPT-5.3 Codex (OpenAI) - typically $1.75 / $14; 400 thousand tokens of context, 22 providers.
- Kimi K2.7 Code (Moonshot) - typically $0.95 / $4, from $0.55 input; open weights, 38 providers.
- Qwen3 Coder 480B (Alibaba) - typically $0.38 / $1.60, from $0.22 input; 19 providers.
- Qwen3 Coder Next - typically $0.20 / $1.20; Qwen3 Coder 30B - about $0.10 / $0.36, the cheapest capable option.
- Seed 2.0 Code (ByteDance) - typically $0.50 / $3; Grok Code Fast 1 (xAI) - typically $0.20 / $1.50.
- GPT-5.1 Codex mini - typically $0.25 / $2, the small OpenAI option.
GPT-5.3 Codex: every provider
| Provider | Model | Input, $ | Output, $ |
|---|---|---|---|
| cometapi | GPT-5.3 Codex | $1.40 | $11.20 |
| poe | GPT-5.3 Codex | $1.59 | $12.73 |
| jiekou | gpt-5.3-codex | $1.66 | $13.30 |
| openai | GPT-5.3-Codex | $1.75 | $14.00 |
| abacus | GPT-5.3 Codex | $1.75 | $14.00 |
| aihubmix | GPT-5.3-Codex | $1.75 | $14.00 |
| azure | GPT-5.3 Codex | $1.75 | $14.00 |
| digitalocean | GPT-5.3 Codex | $1.75 | $14.00 |
| edenai | GPT-5.3 Codex | $1.75 | $14.00 |
| edenai | GPT-5.3 Codex | $1.75 | $14.00 |
| fastrouter | OpenAI: GPT-5.3-Codex | $1.75 | $14.00 |
| freemodel | GPT-5.3 Codex | $1.75 | $14.00 |
Live data: this table is built from today's prices.
Kimi K2.7 Code: every provider
| Provider | Model | Input, $ | Output, $ |
|---|---|---|---|
| kenari | Kimi K2.7 Code | $0.55 | $2.76 |
| inceptron | Kimi K2.7 Code | $0.60 | $3.50 |
| kilo | MoonshotAI: Kimi K2.7 Code | $0.60 | $3.50 |
| openrouter | MoonshotAI: Kimi K2.7 Code | $0.60 | $3.50 |
| melious | Kimi K2.7 Code | $0.61 | $3.29 |
| edenai | Kimi K2.7 Code | $0.62 | $2.60 |
| cortecs | Kimi K2.7 Code | $0.66 | $3.31 |
| deepinfra | Kimi K2.7 Code | $0.68 | $3.40 |
| edenai | Kimi K2.7 Code | $0.68 | $3.40 |
| huggingface | Kimi K2.7 Code | $0.68 | $3.40 |
| iteracompute | Kimi K2.7 Code | $0.69 | $3.29 |
| trustedrouter | MoonshotAI: Kimi K2.7 Code | $0.69 | $3.48 |
Live data: this table is built from today's prices.
Qwen3 Coder 480B: every provider
| Provider | Model | Input, $ | Output, $ |
|---|---|---|---|
| google-vertex | Qwen3-Coder 480B-A35B Instruct | $0.22 | $1.80 |
| helicone | Qwen3 Coder 480B A35B Instruct Turbo | $0.22 | $0.95 |
| vercel | Qwen3 Coder 480B A35B Instruct | $0.22 | $1.80 |
| cometapi | Qwen3-Coder 480B-A35B Instruct | $0.24 | $0.96 |
| requesty | Qwen3-Coder 480B-A35B Instruct | $0.24 | $0.80 |
| siliconflow | Qwen/Qwen3-Coder-480B-A35B-Instruct | $0.25 | $1.00 |
| deepinfra | Qwen3 Coder 480B A35B Instruct Turbo | $0.30 | $1.00 |
| edenai | Qwen3-Coder 480B-A35B Instruct | $0.30 | $1.00 |
| fastrouter | Qwen: Qwen3 Coder 480B A35B Instruct Turbo | $0.30 | $1.00 |
| kilo | Qwen: Qwen3 Coder 480B A35B | $0.30 | $1.00 |
| openrouter | Qwen: Qwen3 Coder 480B A35B | $0.30 | $1.00 |
| requesty | Qwen3-Coder 480B-A35B Instruct | $0.30 | $1.00 |
Live data: this table is built from today's prices.
How to choose
- Hard tasks across a large codebase: GPT-5.3 Codex or Kimi K2.7 Code.
- Agents that run all day: Qwen3 Coder 480B or Coder Next - a fraction of the price, open weights, many hosts.
- Autocomplete and small edits: Qwen3 Coder 30B, Grok Code Fast or GPT-5.1 Codex mini.
- Output is the big cost: agents write long diffs, so compare the output price first; cached input lowers the cost of re-reading the same files. For a monthly estimate with your own token counts use the cost calculator.
All prices in US dollars per 1 million tokens, before taxes, as the providers publish them; every row links to a card with its source and check date.