Featherless is an LLM inference API and chat subscription for open-weight models, used by roleplay writers and developers and billed flat or by credits.
Featherless gives people and applications access to open-weight models without renting, provisioning or configuring a GPU. Open models are plentiful, yet most are hard to run reliably outside a research lab. Featherless hosts them as a serverless service, so a model is ready to query the moment it is picked. The San Francisco company behind it began as Recursal.AI and has carried the Featherless name since 2024. Requests go through a single OpenAI-compatible API covering chat completions, text completions, embeddings, tokenization and a classifier endpoint. Vision-capable models take images by URL or base64, and a text-to-speech endpoint adds voice cloning. Function calling runs natively on the Kimi K2 and Qwen 3 families, with a prompt-based workaround for other models. Thinking models expose an enable_thinking switch for harder problems. Prompts and completions sent through the API are not logged on any plan. The Chat plan is meant for people writing roleplay, stories and long conversations by hand in front ends such as SillyTavern, RisuAI and Wyvern. It is a flat monthly subscription with unlimited tokens, and its terms keep out API traffic, automation and reselling. Developers building coding agents or production applications choose the Developer plan, which draws each request from a monthly credit balance at that model's own token rate. Teams needing guaranteed latency, named engineers and an SLA sign a contract for the Business tier on dedicated GPUs. A common job is pointing a coding agent such as Cline, Kilo Code or OpenHands at an open model rather than a closed one. Automation builders call models from n8n workflows, while application teams reach them through LangChain, LlamaIndex or LiteLLM. Retrieval projects pair the embeddings endpoint with a chat model from the same catalogue, both called with one API key. Featherless suits builders who want to test many open models from a single account and keep the winners running there in production.
Pricing
View plans-
Chat
$25 per month
$25 a month, flat monthly rate with no annual option listed
- Unlimited tokens with any model in the catalogue
- 32K context window and 4 concurrent units
- Works with SillyTavern, RisuAI and Wyvern
- Human-typed interactive chat only; no app or API traffic, reselling, background automation or benchmarking
- Support moves a Chat subscriber onto Developer and pro-rates what was paid
-
Developer
$50 per month
$50 a month credit amount at the selector default ($50 per unit), billed monthly with no annual option listed
- Larger monthly credit amounts can be selected
- Billed per token at each model's own rate, drawn from the credit balance
- Cached input billed below the input rate on some models
- Context up to 256K, 100 concurrent units and priority capacity
- 1 agent environment included
- Credits never expire; one-time top-ups of any amount from Billing
- Only successful requests are charged; requests stop at zero credits until a top-up
- Works with Cline, Pi Agent, Kilo Code, OpenCode, OpenHands and n8n
- Reselling access not allowed
-
Business (Dedicated GPUs)
Custom pricing
- Custom contract price after a talk with an engineer
- Dedicated H100, MI325X, B200 or B300 GPUs, concurrency sized to them
- No shared rate limits; burst to public cloud
- Hosting of private fine-tuned models
- Named engineers and an SLA
- Reselling negotiable
Features
- Included: Published context window size
- Included: Image input support
- Not included: Audio input support
- Included: Function and tool calling
- Not included: JSON schema output mode
- Included: Extended reasoning mode
- Included: Prompt caching discount
- Not included: Batch inference API
- Not included: Fine-tuning offered
- Not included: Open weights release
- Included: Zero data retention option
- Not included: Regional data residency
- Not included: Published safety evaluations
- Not included: Free tier or trial credits
- Included: Enterprise agreement available
- Not included: Published deprecation policy
- Not included: Published benchmark results
- Included: Embeddings endpoint
Additional Features
- OpenAI-compatible API with one key for the whole catalogue
- Chat completions and text completions endpoints
- Embeddings endpoint for catalogue embedding models
- Tokenize and classifier endpoints
- Text-to-speech endpoint with voice cloning
- Vision input by image URL or base64
- Native function calling on Kimi K2 and the Qwen 3 family
- enable_thinking switch for thinking models
Best for
- Writers running roleplay and long stories in SillyTavern, RisuAI or Wyvern
- Chat users who type by hand rather than automate requests
- Developers pointing coding agents such as Cline or OpenHands at open models
- Teams putting open-weight models behind a production app through an OpenAI-style API
- Builders who want to try many open models from one account
- n8n users adding open models to workflow automations
- Companies that need dedicated GPUs, an SLA and named engineers
- Users who want prompts and completions left unlogged
Use cases
- Running roleplay and interactive fiction on an open model
- Writing long stories through a chat front end
- Powering a coding agent with an open-weight model
- Serving production API traffic from open models
- Trying several open models before choosing one for an app
- Generating embeddings for retrieval and search
- Describing images with a vision model
- Returning structured calls to your own tools
- Adding spoken output with text-to-speech and voice cloning
- Building n8n automations that call an open model
- Running open models on reserved GPUs under an SLA
Screenshots & Videos
Explore Featherless in action
User Reviews
Compare with Similar Software
Click to compare side by side with Featherless
