Fireworks AI
Hosted inference and training platform for open models: prepaid pay-per-token serverless, reserved throughput, dedicated GPUs and fine-tuning, with OpenAI- and Anthropic-compatible APIs.
Tags
Description
Commercial inference/training API (fireworks.ai). Established platform, not presented as a brand-new launch. https://www.fireworks.ai 308s to https://fireworks.ai/; use the apex host. Docs at https://docs.fireworks.ai/ (200; also 200 without the trailing slash). Pricing page verified 2026-11-16 (https://fireworks.ai/pricing, 200): usage draws down prepaid credits. Serverless is pay-per-token, advertised from $0.05 / 1M input tokens, with Standard, Priority, Fast, Batch (50% off input and output for jobs that do not need an immediate answer), Spot (30% off Standard on best-effort capacity for select models; throughput can drop to zero), and Ultra-fast (preview, talk to us). Cached input is billed at half price or less, with no write fee. Reserved Throughput is documented as reserve dollars per minute on serverless rates with an SLA. Dedicated Training API is advertised from $8.00 / GPU-hour (the page lists H100 80 GB and H200 141 GB at $8.00/hour). Serverless Training is billed for tokens you prefill, sample and train (per-model table on the pricing page; a previous “from $0.50 / 1M training tokens” floor was not found on the 2026-11-16 page and is not repeated). The live pricing page is authoritative; rates change. Billing FAQs (200): Fireworks uses a usage-based pre-paid credit system (per token for serverless, per GPU time for on-demand deployments, per training token for Serverless Training, per GPU hour for Dedicated Training). If the balance reaches zero and Auto Reload is not enabled, usage pauses until credits are added. A $1 starter credit is documented: without a payment method the account is suspended when that credit is finished; with a payment method you add credits to continue. Contracted customers may move to post-paid billing. Refund terms were not independently extracted beyond those FAQs. No purchase was made. Auth is an API key from the Fireworks dashboard (FIREWORKS_API_KEY; not created). Official Python SDK fireworks-ai, or OpenAI / Anthropic SDKs. OpenAI compatibility (https://docs.fireworks.ai/tools-sdks/openai-compatibility, 200): OpenAI client base URL https://api.fireworks.ai/inference/v1. Anthropic compatibility is documented at base URL https://api.fireworks.ai/inference. Training data can use the OpenAI-compatible chat completion format. Documented API differences include max_tokens shrinking to fit the context window by default (context_length_exceeded_behavior: truncate or error) and usage stats on the last streaming chunk. llms.txt https://docs.fireworks.ai/llms.txt was HTTP 200. license proprietary; pricing paid. No signup or spend was used.
Agent access
Related
- OpenRouter: Unified API for hundreds of LLMs from many providers behind one endpoint and key, with model and price comparison.
- Together AI: Hosted platform to run, train and serve open-source models via prepaid serverless, batch, dedicated and fine-tuning APIs, with an OpenAI-compatible base URL.
Guides & comparisons
- GPU clouds and serverless inference for agents (Comparison): Compare GPU rental, serverless GPUs and hosted per-token APIs for agents: OpenAI-compatible base URLs, tools and structured output, BYO models, scale-to-zero and billing, APIs/CLIs, and host-trust notes. No cached dollar figures or invented benchmarks.
Sources
- https://fireworks.ai
- https://fireworks.ai/pricing
- https://docs.fireworks.ai/
- https://docs.fireworks.ai/llms.txt
- https://docs.fireworks.ai/getting-started/quickstart
- https://docs.fireworks.ai/tools-sdks/openai-compatibility
- https://docs.fireworks.ai/faq-new/billing-pricing/how-much-does-fireworks-cost
- https://docs.fireworks.ai/faq-new/billing-pricing/how-does-billing-and-credit-usage-work
- https://docs.fireworks.ai/faq-new/billing-pricing/what-happens-when-i-finish-my-1-dollar-credit
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)