Open models.Up to 60% below OpenRouter.
DeepSeek, Qwen, GLM, Kimi, MiniMax and gpt-oss behind one OpenAI-compatible endpoint. You swap the base URL and the key and keep the rest of your code.
| Model | OpenRouter | Fluence | Save |
|---|---|---|---|
| gpt-oss-120bOpenAI | in$0.037out$0.17 | in$0.015out$0.068−60% | −60% |
| GLM 5.3 FlashZ.ai | in$0.075out$0.25 | in$0.03out$0.10−60% | −60% |
| Qwen3.6 27BQwen | in$0.30out$2.00 | in$0.135out$0.90−55% | −55% |
Pay less for the models you already run.
Each price is a provider's posted offer for the same checkpoint, checked against OpenRouter's public rate. Zero data retention routes carry their own rates.
| Model | OpenRouter | Fluence | You save | Context |
|---|---|---|---|---|
| gpt-oss-120bopenai/gpt-oss-120b−60% | in$0.037out$0.17 | in$0.015out$0.068 | −60% | |
| GLM 5.3 Flashzai-org/GLM-5.3-Flash−60% | in$0.075out$0.25 | in$0.03out$0.10 | −60% | |
| Qwen3.6 27BQwen/Qwen3.6-27B−55% | in$0.30out$2.00 | in$0.135out$0.90 | −55% | |
| GLM 5zai-org/GLM-5−52% | in$0.60out$1.92 | in$0.288out$0.922 | −52% | |
| DeepSeek V4 Flash 0423deepseek-ai/DeepSeek-V4-Flash−45% | in$0.089out$0.177 | in$0.049out$0.097 | −45% | |
| GLM 4.5 Airzai-org/GLM-4.5-Air−45% | in$0.13out$0.85 | in$0.072out$0.468 | −45% | |
| GLM 4.7zai-org/GLM-4.7−45% | in$0.40out$1.75 | in$0.22out$0.963 | −45% | |
| GLM 4.5zai-org/GLM-4.5−45% | in$0.60out$2.20 | in$0.33out$1.21 | −45% |
Running in production
- Max discount
- 60%
- The largest saving on the list today. Each model on it costs less than on OpenRouter.
- Time to first token
- 2.4s
- You wait as long as you would calling the provider yourself.
- Uptime
- 99.98%
- We move your request to the next provider the moment one drops, at any hour.
Change two values in your client and you are live.
client = OpenAI(base_url="https://api.fluence.cloud/v1", api_key=os.environ["FLUENCE_API_KEY"])- 1
Point your client at Fluence
Set the base URL to api.fluence.cloud/v1 in any OpenAI-compatible SDK, or in curl.
- 2
Paste one key
One key covers all 20 models, and you call each model by its Hugging Face checkpoint id.
- 3
Pay per token
You add funds from $5 and pay per token, with no monthly plan. Each request shows what it cost.
The discount comes from idle GPUs competing for your requests.
GPU operators serve open weights
Teams with idle H100s and H200s host the checkpoints you use today and list capacity below OpenRouter's rate.
We route to the cheapest match
We route your request to the cheapest offer that matches the model and the privacy setting you chose.
OpenRouter's rate is the ceiling
If no offer beats it, you pay OpenRouter's public rate, and that is the most you can pay.
The only record we keep
For each request we store this entry and nothing else.
- Model, token counts, cost and time
- Nothing from your prompts or completions
- Zero data retention routes for policies that require them
- model
- deepseek-ai/DeepSeek-V4-Flash-0731
- input_tokens
- 1,208
- output_tokens
- 342
- cost_usd
- 0.000084
- route
- standard
- created_at
- 2026-09-08T14:02:11Z
- prompt
- not stored
- completion
- not stored
We pay for idle GPUs and spare API capacity.
You serve open models per request and keep your own hardware, accounts and credentials.
Is it the same model?
Yes. Each model on the list is an open-weight checkpoint on Hugging Face, and the id you call names that exact checkpoint. A provider with spare GPUs runs it unchanged.
Which models are on the list?
20 open-weight models with a public rate on OpenRouter: DeepSeek V4, Qwen 3.5 to 3.8, GLM 4.5 to 5.3, Kimi K3, MiniMax M2.7 and gpt-oss-120b. We leave closed models off the list.
What changes in my code?
You change two values: the base URL (api.fluence.cloud/v1) and your API key. Requests and responses follow the OpenAI format, so your SDKs, agents and tools keep working.
How are prices set?
Providers post offers below OpenRouter's public rate, and your request takes the best one that matches. Prices move with supply and stop at that rate.
What do you keep from my requests?
We store the model, the token counts, the cost and the time, and none of the prompt or completion text. Pick a zero data retention route when your policy requires it.
When do I get access?
We send invites in waves, in signup order, and you get one email when yours is ready.
