What does each model cost?
Every rate we hold, with the source named on each row. Six labs publish a pricing page we can read directly; where one does not, the number comes from a licensed index or from two independent registries — and where those two disagree, or only one has the model at all, the row is marked * rather than withheld. Looking for what a fixed level of capability costs rather than what a model costs? That is the AI deflation curve.
27 of 828 rates, 6 marked * — a single registry's word, or two that disagree. Those are the best available numbers rather than confirmed ones.
| Model | Seller | In /Mtok | Out /Mtok | Cached in | Beyond tokens | Named change |
|---|---|---|---|---|---|---|
| Llama Guard 3 8Bstandard · text | Cloudflare | $0.48 | $0.03 | — | — | |
| Granite 4.0 H Microstandard · text | Cloudflare | $0.02 | $0.11 | — | — | |
| Llama 3.2 1B Instructstandard · text | Cloudflare | $0.03 | $0.20 | — | — | |
| Llama 3.1 8B Instruct fp8standard · text | Cloudflare | $0.15 | $0.29 | — | — | |
| gemma-4-26b-a4b-itstandard · text | Cloudflare | $0.10 | $0.30 | — | — | |
| gpt-oss-20bstandard · text | Cloudflare | $0.20 | $0.30 | — | — | |
| Llama 3.2 3B Instructstandard · text | Cloudflare | $0.05 | $0.34 | — | — | |
| Qwen3 30B A3b fp8standard · text | Cloudflare | $0.05 | $0.34 | — | — | |
| GLM-4.7-Flashstandard · text | Cloudflare | $0.06 | $0.40 | — | — | |
| Glm 5.3 Flash*standard · text | Cloudflare | $0.15 | $0.50 | $0.03 | — | |
| Gemma Sea Lion V4 27B Itstandard · text | Cloudflare | $0.35 | $0.56 | — | — | |
| Mistral Small 3.1 24B Instructstandard · text | Cloudflare | $0.35 | $0.56 | — | — | |
| Llama 3.2 11B Vision Instructstandard · text | Cloudflare | $0.05 | $0.68 | — | — | |
| gpt-oss-120bstandard · text | Cloudflare | $0.35 | $0.75 | — | — | |
| Llama 4 Scout 17B 16E Instructstandard · text | Cloudflare | $0.27 | $0.85 | — | — | |
| Qwen2.5 Coder 32B Instructstandard · text | Cloudflare | $0.66 | $1 | — | — | |
| Qwq 32Bstandard · text | Cloudflare | $0.66 | $1 | — | — | |
| deepseek-v4-flash-0731*standard · text | Cloudflare | $0.44 | $1.32 | $0.01 | — | |
| Nemotron 3 Super 120B*standard · text | Cloudflare | $0.50 | $1.50 | — | — | |
| Llama 3.3 70B Instruct fp8 Faststandard · text | Cloudflare | $0.29 | $2.25 | — | — | |
| qwen3.8-27b*standard · text | Cloudflare | $0.45 | $3.20 | $0.05 | — | |
| deepseek-v4-pro-0813*standard · text | Cloudflare | $1.32 | $3.96 | $0.04 | — | |
| kimi-k2.6standard · text | Cloudflare | $0.95 | $4 | $0.16 | — | |
| kimi-k2.7-codestandard · text | Cloudflare | $0.95 | $4 | $0.19 | — | |
| Glm 5.2standard · text | Cloudflare | $1.40 | $4.40 | $0.26 | — | |
| Glm 5.3*standard · text | Cloudflare | $1.40 | $4.40 | $0.26 | — | |
| Deepseek R1 Distill Qwen 32Bstandard · text | Cloudflare | $0.50 | $4.88 | — | — |
Beyond tokens99 rates in 12 units
Not everything is sold by the token. An image, a second of video, a minute of speech and a million characters are each their own measurement, and a table that answered only in tokens left them out. They are printed here in the unit the seller published them in, and never converted into one another.
per image25 rates
| 200px x 200px | Voyage | $0.00per image |
| 1000px x 1000px | Voyage | $0.00per image |
| 2000px x 2000px | Voyage | $0.00per image |
| 4000px x 4000px | Voyage | $0.00per image |
| CogView-4 | Z.ai | $0.01per image |
| GLM-Image | Z.ai | $0.01per image |
| FLUX.2 [dev] | Together | $0.02per image |
| Gemini 2.5 Flash Image 🍌 | $0.02per image | |
| Gemini 2.5 Flash Image 🍌 | $0.02per image | |
| grok-imagine-image | xAI | $0.02per image |
| FLUX.2 [flex] | Together | $0.03per image |
| FLUX.2 [pro] | Together | $0.03per image |
| Wan 2.6 Image | Together | $0.03per image |
| GPT Image 1.5 | Together | $0.03per image |
showing the cheapest 14 of 25
per video24 rates
| ByteDance Seedance 2.5 | Together | $0.12per video |
| ByteDance Seedance 1.0 Lite | Together | $0.14per video |
| HappyHorse 1.1 I2V | Together | $0.14per video |
| HappyHorse 1.1 R2V | Together | $0.14per video |
| HappyHorse 1.1 T2V | Together | $0.14per video |
| ByteDance Seedance 2.0 | Together | $0.16per video |
| FLUX 3 | Together | $0.17per video |
| Kling 2.1 Standard | Together | $0.18per video |
| Kling 1.6 Standard | Together | $0.19per video |
| CogVideoX-3 | Z.ai | $0.20per video |
| Popular Special Effects Video Templates | Z.ai | $0.20per video |
| Vidu Q1 | Together | $0.22per video |
| HappyHorse 1.0 I2V | Together | $0.24per video |
| HappyHorse 1.0 R2V | Together | $0.24per video |
showing the cheapest 14 of 24
per hour17 rates
| CCU price | Anthropic | $0.01per hour |
| Conversion | Anthropic | $0.01per hour |
| Per GiB RAM | Together | $0.01per hour |
| Per vCPU | Together | $0.04per hour |
| NVIDIA HGX H100 | Together | $1.99per hour |
| NVIDIA HGX H200 | Together | $2.99per hour |
| NVIDIA HGX H100 | Together | $3.69per hour |
| NVIDIA HGX H100 | Together | $3.99per hour |
| NVIDIA HGX H100 | Together | $3.99per hour |
| NVIDIA HGX B200 | Together | $4.09per hour |
| NVIDIA HGX H200 | Together | $4.99per hour |
| NVIDIA HGX H200 | Together | $5.99per hour |
| NVIDIA HGX H200 | Together | $5.99per hour |
| NVIDIA HGX B200 | Together | $7.99per hour |
showing the cheapest 14 of 17
per million characters9 rates
| NVIDIA Parakeet TDT 0.6B V3 Realtime | Together | $0.00per million characters |
| Inkling | Together | $1per million characters |
| Kokoro-82M TTS | Together | $4per million characters |
| Orpheus TTS | Together | $15per million characters |
| Text to Speech | xAI | $15per million characters |
| tts-1 | OpenAI | $15per million characters |
| tts-1-hd | OpenAI | $30per million characters |
| Cartesia Sonic-2 | Together | $65per million characters |
| Cartesia Sonic-3 | Together | $65per million characters |
per request7 rates
| rerank-2.5-lite | Voyage | $0.00per request |
| rerank-3-lite | Voyage | $0.00per request |
| rerank-2.5 | Voyage | $0.00per request |
| rerank-3 | Voyage | $0.00per request |
| Lyria 3 Clip Preview | $0.04per request | |
| Lyria 3 Pro Preview | $0.08per request | |
| Lyria 3.5 | $0.08per request |
per thousand calls5 rates
| Collections Search | xAI | $2.50per thousand calls |
| Code Execution | xAI | $5per thousand calls |
| Web Search | xAI | $5per thousand calls |
| X Search | xAI | $5per thousand calls |
| File Attachments | xAI | $10per thousand calls |
per thousand images4 rates
| 200px x 200px | Voyage | $0.03per thousand images |
| 1000px x 1000px | Voyage | $0.60per thousand images |
| 2000px x 2000px | Voyage | $1.20per thousand images |
| 4000px x 4000px | Voyage | $1.20per thousand images |
per second4 rates
| grok-imagine-video | xAI | $0.05per second |
| sora-2 | OpenAI | $0.05per second |
| grok-imagine-video-1.5 | xAI | $0.08per second |
| sora-2 | OpenAI | $0.10per second |
per billion pixels1 rate
| voyage-multimodal-3.5 voyage-multimodal-3 | Voyage | $0.60per billion pixels |
per minute1 rate
| gpt-live-1 | OpenAI | $0.05per minute |
per session1 rate
| Session runtime | Anthropic | $0.08per session |
per use1 rate
| Web Search | Z.ai | $0.01per use |
† The seller names no unit on that table. It was read from the section it sits under, or from a column named for a token role, and it is marked rather than presented as the seller's own words.
What would it cost to actually do something?
A rate per million tokens is how the industry sells, and almost nobody can feel it. That question — one question, one long document, one day of an agent — has a page of its own, with the same numbers in the unit a person has.
How to read this table
Prices are per million tokens, in US dollars. Every row names the seller it came from, and where that is not the seller's own published page the mark says so: * means the number comes from a single registry or two that disagree, so it is the best available rather than a confirmed one. The same mark covers both because the remedy is the same — check the seller before you commit.
A row is a rate, not a model. One model can appear four times at four prices — standard, batch, flex, fast — and again for audio or image tokens, and again either side of a context threshold. This page opens on the standard text rates because those are the ones a reader compares; the toggle shows everything.
Where a seller has announced a rise, it is named here. A dated increase a reader can still act on is the most useful thing this table holds, so it sits beside the current rate rather than in a footnote.