Verified 2026-11-18 · rates follow the live catalog
A price per million tokens is hard to feel. The 50 billed text models on this page have input prices that differ by 1,000×, and that spread is the whole story of cheap inference. To make it concrete, every price below is turned into the cost of one request: a 20,000-token prompt answered with a 1,000-token reply and nothing cached. That is an illustration of a single coding-agent step, not a measurement of your traffic, so scale it to your own prompt sizes.
| Model | Input / 1M | Output / 1M | One request | 1,000 requests | One request, prompt cached |
|---|---|---|---|---|---|
| Gemma 4 26B | $0.01 | $0.03 | $0.00023 | $0.23 | — |
| GPT-OSS 20B | $0.02 | $0.03 | $0.00043 | $0.43 | — |
| Llama 4 Scout | $0.027 | $0.085 | $0.00062 | $0.62 | — |
| MiniMax M3 | $0.03 | $0.12 | $0.00072 | $0.72 | $0.00024 |
| GPT-OSS 120B | $0.035 | $0.075 | $0.00077 | $0.77 | — |
| Qwen3.8 27b | $0.045 | $0.32 | $0.0012 | $1.22 | — |
| GPT-6 Luna | $0.05 | $0.25 | $0.0013 | $1.25 | $0.00035 |
| Hy4 Flash | $0.063 | $0.2529 | $0.0015 | $1.51 | $0.00031 |
| GLM-4.7-Flash | $0.06 | $0.40 | $0.0016 | $1.60 | — |
| Qwen3.8 Flash | $0.075 | $0.235 | $0.0017 | $1.74 | $0.0004 |
| Claude Haiku 5.5 | $0.10 | $0.50 | $0.0025 | $2.50 | $0.0007 |
| MiMo-V2.5 | $0.14 | $0.28 | $0.0031 | $3.08 | — |
Every figure is what the model bills at the most expensive hour of the day, so it is the most a request can cost. A discount that applies at every hour is already inside it, and scheduled discounts, which lower the price further during set hours, are listed under published discounts below.
Gemma 4 26B is the cheapest at $0.00023 a request, and Claude Fable 5.1 is the dearest at $0.25: 1,087× apart. At 100,000 requests a month that is $23 against $25,000.
| Model | One request | Multiple of the cheapest |
|---|---|---|
| Claude Fable 5.1 | $0.25 | 1,087× |
| Claude Fable 5 | $0.25 | 1,087× |
| Claude Fable 5.2 | $0.20 | 870× |
| GPT-6 Astra | $0.125 | 543× |
These rows are here so the gap is visible, not to argue that the cheap models match them. Price does not measure quality, and a cheaper model that needs two attempts is not cheaper. Shortlist by price, then run the shortlist on your own tasks.
These models publish a discount against their list price:
Each model's page shows its list price, its rates and its schedule.
40 of the 50 models above publish a separate, lower price for input tokens served from cache, which matters for agents that resend a long, stable prefix on every step. On MiMo-V2.6-Pro the same request costs $0.00094 when the whole prompt is a cache hit, against $0.0096 when it is not.
Before paying for anything, check the free models: 2 models are free to call for a limited time.
Agents spend by the session rather than by the request. The cost-per-session guide ranks models by what a whole coding-agent session costs.
On-demand credits bill per token at the rates above. The $20 Minion and $80 Boss plans add an included allowance, and usage past it bills as on-demand credits. If your on-demand spend regularly passes a plan's monthly price, compare the two on the pricing page.
At its most expensive hour of the day, Gemma 4 26B is the cheapest billed text model here, at $0.00023 for the request above. Whether it is good enough for your codebase is the part a price cannot tell you.
Not necessarily. A model that needs more attempts, or a longer prompt to reach the same result, can cost more per finished task. Compare cost per request, then measure cost per completed task on a sample of your own work.
Model catalog · Pricing · Comparisons · Cost guides · Integration guides · FAQ
Guides: GLM coding plan alternative · Free AI API · LLM gateway · LLM proxy · Cheap AI API · OpenAI-compatible API
Model APIs: GLM API · DeepSeek API · Kimi API · Qwen API · MiniMax API