Decentralized AI Inference on Bittensor
Subnets serving AI models (LLMs, image and speech models) through APIs run by a decentralized network of miners, often cheaper than centralized providers.
💡 In plain words
When you ask an AI a question, a computer somewhere has to run the model to answer. These subnets do that job, like ChatGPT's API, but on thousands of independent computers.
🏢 What you already know
🪙 On the crypto side
The difference on Bittensor: nobody owns the service. Independent participants compete to do the job best, and the network pays the best ones in the subnet's token.
-
Chutes 8.97%
-
engy 4.99%
-
SayGM 3.26%
-
Green Compute 2.41%
-
Cacheon 0.64%
| # | Subnet | Price | 24h | 7d | Market cap | Emission | Vol 24h |
|---|---|---|---|---|---|---|---|
| 64 | $19.17 | ▼ 0.55% | ▼ 3.67% | $67.1M | 8.97% | — | |
| 53 | $10.75 | ▼ 0.90% | ▲ 17.58% | $55.4M | 4.99% | — | |
| 28 | $7.85 | ▲ 14.07% | ▲ 33.03% | $42.2M | 3.26% | — | |
| 95 | $9.59 | ▲ 1.59% | ▲ 4.20% | $29.7M | 0.00% | — | |
| 110 | $5.69 | ▲ 5.12% | ▲ 17.79% | $18.6M | 2.41% | — | |
| 14 | $2.40 | ▼ 0.89% | ▼ 1.63% | $7.89M | 0.64% | — | |
| 10 | $1.65 | ▼ 3.17% | ▼ 3.68% | $5.73M | 0.00% | — | |
| 46 | $1.02 | ▼ 2.72% | ▲ 3.50% | $3.68M | 0.00% | — | |
| 2 | $0.8245 | ▲ 0.21% | ▼ 2.13% | $2.79M | 0.00% | — | |
| 65 | $0.6822 | ▲ 8.94% | ▲ 20.21% | $2.48M | 0.00% | — | |
| 29 | $0.7034 | ▼ 0.07% | ▼ 0.19% | $2.14M | 0.00% | — | |
| 96 | $1.29 | ▲ 13.66% | ▲ 5.85% | $1.23M | 0.01% | — | |
| 40 | $0.6271 | ▲ 1.38% | ▼ 5.16% | $324K | 0.01% | — |
Categories are assigned editorially from each project's description and website; they are not stored on-chain.
What is decentralized AI inference?
Decentralized AI inference means running AI models, mostly open-weight LLMs such as DeepSeek, Qwen, GLM or Kimi, on a network of independent GPU operators instead of a single cloud provider. On Bittensor, inference subnets reward their miners in alpha for serving models quickly and reliably, which subsidizes the price paid by users. Several of them sell access through OpenAI-compatible APIs, so existing apps, coding agents and chat front-ends can switch by changing a base URL.
They also differ in how they prove what ran: Chutes runs models inside Trusted Execution Environments (TEE), while engy uses cryptographic verification of the model's weights and activations.
Compare LLM API prices across Bittensor subnets (live)
Input / output price in USD per million tokens, read from each subnet's public API every 10 minutes.
| Model | Chutes (SN64) | engy (SN53) | SayGM (SN28) | Official API (in / out) | Market median (in / out) |
|---|---|---|---|---|---|
| DeepSeek-V4-Flash-0731 | $0.44 / $1.32 | $0.045 / $0.09 cheapest | $0.0866 / $0.2599 | — | $0.132 / $0.528 |
| glm-5.2 | $1.25 / $3.95 | $0.68 / $1.5 | $1.26 / $3.96 | $1.4 / $4.4 | $1.155 / $4.4 |
| kimi-k3 | $1.85 / $9.5 | $1.95 / $9.75 | $2.85 / $14.25 | $3 / $15 | $2.75 / $14.95 |
| qwen3.8-27b | $0.039 / $1.25 cheapest | $0.045 / $0.32 | $0.045 / $0.27 | $0.425 / $2.55 | $0.2 / $2.35 |
| deepseek-v4.1-flash | — | $0.04 / $0.08 cheapest | $0.14 / $0.5599 | $0.15 / $0.6 | $0.205 / $1.07 |
| glm-5.3-flash | — | $0.135 / $0.45 | $0.132 / $0.44 cheapest | $0.15 / $0.5 | $0.1313 / $0.5 |
| glm-5.3-tee | — | $0.98 / $3.08 | $1.26 / $3.96 | $1.4 / $4.4 | $1.225 / $4.4 |
| Mistral-Nemo-Instruct-2407 | $0.0245 / $0.0978 | — | $0.0233 / $0.0929 cheapest | — | — |
| Nemotron-3-Nano-Omni-30B | $0.0245 / $0.0978 | — | $0.0233 / $0.0929 cheapest | — | — |
| qwen3.6-35b-a3b | — | $0.045 / $0.3 cheapest | $0.0771 / $0.4618 | — | $0.125 / $1 |
| Qwen3-32B | $0.104 / $0.416 | — | $0.0988 / $0.3952 cheapest | — | $0.14 / $0.4 |
| gemma-4-31B-turbo | $0.12 / $0.37 | — | $0.114 / $0.3515 cheapest | — | — |
| qwen3.6-27b-tee | $0.25 / $2 cheapest | — | $0.285 / $1.9 | $0.45 / $2.7 | $0.32 / $3.2 |
| Qwen3-235B-A22B-Thinking-2507 | $0.2989 / $1.1957 | — | $0.284 / $1.1359 cheapest | — | $0.45 / $3.5 |
| qwen3.5-397b-a17b-tee | $0.45 / $3 | — | $0.4275 / $2.85 cheapest | $0.39 / $2.34 | $0.55 / $3.55 |
| Kimi-K2.6 | $0.45 / $1.95 cheapest | — | $0.475 / $2.7075 | $0.95 / $4 | $0.75 / $3.5 |
| ornith-1.5-397b | — | — | $0.7 / $2.2 cheapest | — | — |
| deepseek-v3.2-tee | $1 / $1 | — | $0.95 / $0.95 cheapest | — | $0.29 / $0.88 |
| glm-5.1-tee | $0.98 / $3.08 cheapest | — | $1.33 / $4.18 | $1.4 / $4.4 | $1.355 / $4.3 |
| mimo-v2.6-pro-ultraspeed | — | — | $3.915 / $7.83 cheapest | — | $4.35 / $8.7 |
| veo-3.1-fast | — | — | — / — | — | — |
| claude-fable-5-2 | — | — | — / — | — | — |
| h3 | — | — | — / — | — | — |
| wan-2.6-t2v | — | — | — / — | — | — |
| flux.2-klein-4b | — | — | — / — | — | — |
| gpt-oss-20b | — | — | $0.0665 / $0.285 | — | $0.03 / $0.15 |
| gpt-6-luna | — | — | $0.0745 / $0.3725 | — | $0.11 / $0.55 |
| claude-haiku-5-5 | — | — | $0.085 / $0.425 | — | — |
| gpt-5.4-nano | — | — | $0.149 / $0.9313 | — | $0.2 / $1.25 |
| gpt-5.6-luna | — | — | $0.149 / $0.894 | — | $0.22 / $1.32 |
| gemini-3.1-flash-lite | — | — | $0.1625 / $0.975 | $0.125 / $0.75 | $0.2625 / $1.575 |
| gemini-3.1-flash-lite-image | — | — | $0.1625 / $0.975 | $0.25 / $1.5 | $0.25 / $1.5 |
| gemini-3.5-flash-lite | — | — | $0.195 / $1.625 | $0.15 / $1.25 | $0.315 / $2.625 |
| gemini-3.1-flash-image | — | — | $0.325 / $1.95 | $0.5 / $3 | $0.5 / $3 |
| gemini-3.8-flash | — | — | $0.4125 / $2.0625 | $0.375 / $1.875 | $0.75 / $3.75 |
| gemini-3.6-flash | — | — | $0.4875 / $2.4375 | $0.375 / $1.875 | $0.75 / $3.75 |
| gemini-3.7-flash | — | — | $0.4875 / $2.4375 | $0.375 / $1.875 | $0.75 / $3.75 |
| gpt-5.4-mini | — | — | $0.5588 / $3.3525 | — | $0.75 / $4.5 |
| o4-mini | — | — | $0.8195 / $3.278 | — | $1.1 / $4.4 |
| claude-haiku-4-5 | — | — | $0.824 / $4.12 | — | — |
| gemini-3.5-flash | — | — | $0.975 / $5.85 | $0.75 / $4.5 | $1.5 / $9 |
| gemini-3.1-pro-preview | — | — | $1.3 / $7.8 | $1 / $6 | $2 / $12 |
| gpt-5.6-terra | — | — | $1.49 / $8.94 | — | $2.2 / $13.2 |
| o3 | — | — | $1.49 / $5.96 | — | $2 / $8 |
| gpt-6-sol | — | — | $1.49 / $7.45 | — | $2.2 / $11 |
| gpt-6.1-sol | — | — | $1.49 / $7.45 | — | $2.2 / $11 |
| claude-sonnet-5 | — | — | $1.648 / $8.24 | — | $2.1 / $10.5 |
| claude-sonnet-5-5 | — | — | $1.736 / $8.68 | — | — |
| gpt-5.4 | — | — | $1.8625 / $11.175 | — | $2.75 / $16.5 |
| claude-sonnet-4-6 | — | — | $2.472 / $12.36 | — | — |
| gpt-5.6-sol | — | — | $2.98 / $14.9 | — | $4 / $20 |
| claude-opus-5-5 | — | — | $3.472 / $17.36 | — | — |
| gpt-5.5 | — | — | $3.725 / $22.35 | — | $5.5 / $33 |
| claude-opus-4-7 | — | — | $4.12 / $20.6 | — | — |
| claude-opus-4-8 | — | — | $4.12 / $20.6 | — | — |
| claude-opus-5 | — | — | $4.12 / $20.6 | — | $5.5 / $27.5 |
| gpt-6-astra | — | — | $7.45 / $37.25 | — | $11 / $55 |
| claude-fable-5 | — | — | $8.24 / $41.2 | — | $10 / $50 |
| claude-fable-5-1 | — | — | $8.24 / $41.2 | — | — |
| gpt-5.5-pro | — | — | $22.35 / $134.1 | — | $22.5 / $135 |
Off-chain data from the providers' public APIs (llm.chutes.ai, api.engy.ai, api.saygm.com); official API and market median from centralized providers on OpenRouter. “cheapest” compares input prices when a model is served by several subnets.