Decide
Model pricing
What each model charges per million tokens, copied from the vendor's own price list. Every row carries the date the figure was last checked and the page it came from.
Last checked: · 42 items
| Model | Vendor | Input / 1M | Output / 1M | Cached | Task cost | Checked |
|---|---|---|---|---|---|---|
| Claude Fable 5claude-fable-5 | Anthropic | $10.00 | $50.00 | $1.00 | $0.75 | August 22 |
| Claude Mythos 5claude-mythos-5preview | Anthropic | $10.00 | $50.00 | $1.00 | $0.75 | August 22 |
| Claude Opus 5claude-opus-5 | Anthropic | $5.00 | $25.00 | $0.50 | $0.375 | August 22 |
| Claude Opus 4.8claude-opus-4-8 | Anthropic | $5.00 | $25.00 | $0.50 | $0.375 | August 22 |
| Claude Opus 4.7claude-opus-4-7 | Anthropic | $5.00 | $25.00 | $0.50 | $0.375 | August 22 |
| Claude Opus 4.6claude-opus-4-6 | Anthropic | $5.00 | $25.00 | $0.50 | $0.375 | August 22 |
| Claude Opus 4.5claude-opus-4-5 | Anthropic | $5.00 | $25.00 | $0.50 | $0.375 | August 22 |
| Claude Sonnet 5claude-sonnet-5 | Anthropic | $2.00 | $10.00 | $0.20 | $0.15 | August 22 |
| Claude Sonnet 4.6claude-sonnet-4-6 | Anthropic | $3.00 | $15.00 | $0.30 | $0.225 | August 22 |
| Claude Sonnet 4.5claude-sonnet-4-5 | Anthropic | $3.00 | $15.00 | $0.30 | $0.225 | August 22 |
| Claude Haiku 4.5claude-haiku-4-5 | Anthropic | $1.00 | $5.00 | $0.10 | $0.075 | August 22 |
| GPT-5.6 Solgpt-5.6-sol | OpenAI | $4.00 | $20.00 | $0.40 | $0.30 | August 22 |
| GPT-5.6 Terragpt-5.6-terra | OpenAI | $2.00 | $12.00 | $0.20 | $0.16 | August 22 |
| GPT-5.6 Lunagpt-5.6-luna | OpenAI | $0.20 | $1.20 | $0.02 | $0.016 | August 22 |
| GPT-5.6 Cybergpt-5.6-cyber | OpenAI | $12.50 | $75.00 | $1.25 | $1.00 | August 22 |
| Gemini 3.1 Progemini-3.1-pro-previewpreview | $2.00 | $12.00 | $0.20 | $0.16 | August 22 | |
| Gemini 3.7 Flashgemini-3.7-flash | $0.75 | $3.75 | $0.075 | $0.0563 | August 22 | |
| Gemini 3.6 Flashgemini-3.6-flash | $0.75 | $3.75 | $0.075 | $0.0563 | August 22 | |
| Gemini 3 Flashgemini-3-flash-previewpreview | $0.50 | $3.00 | $0.05 | $0.04 | August 22 | |
| Gemini 2.5 Progemini-2.5-pro | $1.25 | $10.00 | $0.125 | $0.1125 | August 22 | |
| Gemini 2.5 Flashgemini-2.5-flash | $0.30 | $2.50 | $0.03 | $0.0275 | August 22 | |
| Claude Opus 4.1claude-opus-4-1retired | Anthropic | $15.00 | $75.00 | $1.50 | $1.125 | August 22 |
| Claude Sonnet 4claude-sonnet-4retired | Anthropic | $3.00 | $15.00 | $0.30 | $0.225 | August 22 |
| Claude Haiku 3.5claude-haiku-3-5retired | Anthropic | $0.80 | $4.00 | $0.08 | $0.06 | August 22 |
Task cost is calculated for 50,000 input and 5,000 output tokens. Change the numbers above the table.
Open-weight models
The same model, several prices. The weights can be downloaded, so it is served both by the lab that made it and by anyone renting machines — and the gap runs to several times. Every row therefore says who is charging, not only who built the model.
| Model | Built by | Served by | Input / 1M | Output / 1M | Cached | Task cost | Checked |
|---|---|---|---|---|---|---|---|
| DeepSeek V4 Flashdeepseek-v4-flash | DeepSeek | the maker itself | $0.44 | $1.32 | $0.014 | $0.0286 | August 26 |
| DeepSeek V4 Prodeepseek-v4-pro | DeepSeek | the maker itself | $1.32 | $3.96 | $0.044 | $0.0858 | August 26 |
| DeepSeek V4 Flash 0731deepseek/deepseek-v4-flash-0731 | DeepSeek | OpenRouter | $0.06 | $0.12 | — | $0.0036 | August 26 |
| DeepSeek V4 Pro 0813deepseek/deepseek-v4-pro-0813 | DeepSeek | OpenRouter | $1.122 | $3.366 | — | $0.0729 | August 26 |
| Qwen3.8 27Bqwen/qwen3.8-27b | Alibaba | OpenRouter | $0.425 | $2.55 | — | $0.034 | August 26 |
| Kimi K2.7 Codemoonshotai/kimi-k2.7-code | Moonshot AI | OpenRouter | $0.67 | $3.40 | — | $0.0505 | August 26 |
| MiniMax M3minimax/minimax-m3 | MiniMax | OpenRouter | $0.30 | $1.20 | — | $0.021 | August 26 |
| IBM Granite 4.1 8Bibm-granite/granite-4.1-8b | IBM | OpenRouter | $0.05 | $0.10 | — | $0.003 | August 26 |
| gpt-oss-120bopenai/gpt-oss-120b | OpenAI | OpenRouter | $0.037 | $0.17 | — | $0.0027 | August 26 |
| Llama 4 Maverickmeta-llama/llama-4-maverick | Meta | OpenRouter | $0.20 | $0.80 | — | $0.014 | August 26 |
Downloading the weights costs nothing; this is the price when someone else runs the model for you. What it costs to run it yourself is a question about the machine, not about tokens.
What the number does not say
- Claude Fable 5
- Uses the newer tokenizer, which produces about 30% more tokens for the same text than models up to Sonnet 4.6. Price per token is therefore not directly comparable with older models.
- Claude Mythos 5
- Limited availability. Same tokenizer as Fable 5.
- Claude Sonnet 5
- The introductory $2 and $10 pricing is now the standard price. The increase to $3 and $15 scheduled for September 1, 2026 will not happen.
- GPT-5.6 Sol
- Promotional price from August 21 through at least November 21, 2026; before that it was $5 and $30. Listed for short context; long context is $8 and $30 per million. Regional processing adds 10% for models released on or after March 5, 2026.
- GPT-5.6 Terra
- Short context. Long context is $4 and $18 per million.
- GPT-5.6 Luna
- Short context. Long context is $0.40 and $1.80 per million.
- GPT-5.6 Cyber
- From the Daybreak program. No long-context pricing.
- Gemini 3.1 Pro
- Above 200k tokens in the prompt the price is $4 and $18 per million, and cache reads are $0.40 instead of $0.20. Cache storage is $4.50 per million tokens per hour. No free tier.
- Gemini 3.7 Flash
- This price holds through December 31, 2026. From January 1, 2027 it doubles to $1.50 and $7.50 per million. The cache doubles too: reads go from $0.075 to $0.15, and storage from $0.50 to $1.00 per million tokens per hour.
- Gemini 3.6 Flash
- Same notice as 3.7 Flash: from January 1, 2027 the price is $1.50 and $7.50 per million, with the same doubling on the cache. On price it is identical to 3.7 Flash today.
- Gemini 3 Flash
- The price covers text, image and video input; audio is $1 per million. Context storage is billed separately at $1 per million tokens per hour.
- Gemini 2.5 Pro
- Above 200k tokens in the prompt: $2.50 and $15 per million.
- Gemini 2.5 Flash
- The price covers text, image and video input; audio is $1 per million. Cache storage is billed separately at $1 per million tokens per hour. No price change is announced.
- Nano Banana Pro
- The price covers 1K and 2K images; a 4K image is $0.24. Billed by tokens, at $120 per million image output tokens.
- Imagen 4 Standard
- Fast is $0.02 and Ultra is $0.06 per image.
- GPT Image 2
- Billed by tokens rather than per image: $8 input and $30 output per million image tokens. Text input is $5 per million.
- Veo 3.1 Standard
- With audio, at 720p and 1080p; 4K is $0.60 per second. Paid tier only.
- Veo 3.1 Fast
- At 720p; 1080p is $0.12 and 4K is $0.30 per second.
- Veo 3.1 Lite
- At 720p; 1080p is $0.08 per second. No 4K output.
- Sora 2
- At 720p, portrait and landscape.
- Sora 2 Pro
- At 720p; 1024p is $0.50 and 1080p is $0.70 per second.
- DeepSeek V4 Flash
- This is the peak rate. Off-peak is exactly half — 0.22 and 0.66 — and peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday. Context is one million tokens.
- DeepSeek V4 Pro
- Peak rate; off-peak is half. A cache hit costs thirty times less than input read afresh, so on long conversations the cache decides the bill.
- DeepSeek V4 Flash 0731
- The same model at seven times cheaper input than the maker charges, because a third party serves it from its own machines. What the price does not show is exactly who, at what speed, and under whose data handling.
- Qwen3.8 27B
- Twenty-seven billion parameters is a size that fits in some thirty gigabytes at four bits, which is a machine you can buy. This price is for when you would rather not.
- IBM Granite 4.1 8B
- Eight billion parameters under Apache 2.0. Models this small are usually run locally; the per-token price here shows how little it costs when you do not.
- gpt-oss-120b
- OpenAI's open-weight model, and the cheapest input in this table. It is also one of the three models OpenAI used to benchmark its own Jalapeño chip.
- Llama 4 Maverick
- Meta's licence is neither Apache nor MIT but its own, with terms that depend on how many users the product embedding it has. Anyone shipping it should read the licence, not only the price.
- Claude Opus 4.1
- Retired, except on Bedrock and Google Cloud.
- Claude Sonnet 4
- Retired, except on Bedrock and Google Cloud.
- Claude Haiku 3.5
- Retired, except on Bedrock and Google Cloud.
Image, video and audio
These are not billed per token, so they cannot share a column. The unit is next to the price.
Nano Banana Pro
$0.134 per image
The price covers 1K and 2K images; a 4K image is $0.24. Billed by tokens, at $120 per million image output tokens.
Imagen 4 Standard
$0.040 per image
Fast is $0.02 and Ultra is $0.06 per image.
GPT Image 2
$8.00 / $30.00 per million
Billed by tokens rather than per image: $8 input and $30 output per million image tokens. Text input is $5 per million.
Veo 3.1 Standard
$0.40 per second
With audio, at 720p and 1080p; 4K is $0.60 per second. Paid tier only.
Veo 3.1 Fast
$0.10 per second
At 720p; 1080p is $0.12 and 4K is $0.30 per second.
Veo 3.1 Lite
$0.05 per second
At 720p; 1080p is $0.08 per second. No 4K output.
Sora 2
$0.10 per second
At 720p, portrait and landscape.
Sora 2 Pro
$0.30 per second
At 720p; 1024p is $0.50 and 1080p is $0.70 per second.
Prices change without notice. If you find a difference between this and the price list, the price list is right — and tell us, so we can fix it the same day.