Antimatter
A unified LLM gateway giving one API key access to hundreds of AI models over OpenAI, Anthropic and Responses wire formats.
At a Glance
Engagement
Available On
Alternatives
Listed Oct 2026
About Antimatter
Antimatter is an API gateway that exposes a catalog of AI models from many vendors behind one key and one base URL. Developers can connect an existing OpenAI or Anthropic client, pick a model by its vendor/model id, and send requests, with routing, fallback, evaluation and agent tooling layered on top. The homepage describes 474 models behind one key and lets people try models free without signing up.
What It Is
Antimatter is a model gateway and router. It speaks the OpenAI chat wire, the Anthropic messages wire and the OpenAI Responses wire, so official SDKs work by changing the base URL and key. The docs describe switching from OpenRouter as a two-line change, since model ids use the same vendor/model format. Beyond chat completions it serves embeddings, images, text to speech, transcription, batches, and a realtime voice WebSocket.
Routing and Model Selection
The antimatter/auto ids are autopilot policies that pick a model for you: balanced, cheapest, fastest, code, extraction, reasoning, long context, vision, tool calling and structured output. The docs say task modes rely on measured capability probes and small scheduled task probes. Callers can also pass a list of models for fallback on 429s, 5xx errors or dropped connections, or opt into comparable-model fallback. Ensembles send one request to 2 to 8 models and combine answers by rules such as majority or JSON consensus.
Testing and Optimization Tools
User-defined eval suites (exact, contains, regex, JSON schema or number checks) report each model's pass rate, cost and latency, and the cheapest passing model can be routed to with an antimatter/auto:suite id. Shadow tests replay a sampled share of captured traffic to candidate models and report agreement with confidence intervals. Optional request compression, strict JSON repair and response caching are available through headers.
Agents and Integrations
Agent wallets give an agent its own sub-balance, key and spend policy, and hosted agent runs can use function, code interpreter, MCP and image generation tools. A hosted MCP server at /mcp exposes tools for the catalog, cost estimates, chat, ensembles and agent management, with Claude Code and Cursor setup shown. The docs also cover vector stores with file search, long-term memory spaces, a paid MCP tools market, and verifiable usage receipts. Several features, such as video, private inference and distillation, are described as not enabled on this deployment. Credits are topped up with the $ANTI token, per the docs.
Community Discussions
Be the first to start a conversation about Antimatter
Share your experience with Antimatter, ask questions, or help others learn from your insights.
Pricing
Pay-as-you-go (upstream price + 10%)
Metered per-token access to 474 models through one key, billed at the upstream's published price plus a 10% margin. Prices per 1M tokens.
- Margin: +10% over upstream's published price on every model
- No fee to add credit
- Gemini 4 Argon: $2.20 in / $11.00 out per 1M tokens
- Claude Haiku 5.5: $0.11 in / $0.55 out per 1M tokens
- Nano Banana 2.1: $1.65 in / $8.25 out per 1M tokens
- Mistral Large 4: $1.496 in / $4.598 out per 1M tokens
- GPT-6.1 Sol Pro: $2.20 in / $11.00 out per 1M tokens
Capabilities
Key Features
- Unified API for hundreds of models with one key
- OpenAI chat, Anthropic messages and OpenAI Responses wire compatibility
- Autopilot routing policies (cheapest, fastest, code, reasoning, vision, tools, JSON)
- Model fallback chains and comparable-model fallback
- Ensembles with majority, JSON consensus, first-valid and judge rules
- Custom eval suites to route to the cheapest passing model
- Shadow tests on captured traffic
- Request compression and strict JSON repair
- Streaming with resumable streams
- Agent wallets with spend policies and hosted agent runs
- Hosted MCP server and paid MCP tools market
- Vector stores with file search
- Long-term memory spaces
- Realtime voice over WebSocket
- Verifiable Merkle-proof usage receipts
- Bring your own provider key
