Meta Model API
Listed at https://developer.meta.com/ai/
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | $1.25/1M tokens | per million tokens |
| Output | $4.25/1M tokens | per million tokens |
| Cache Read | $0.15/1M tokens (cached input) | Discounted |
🤖 Supported Models (1)
✨ Pros
- ✓Meta's official AI API, Muse Spark 1.1 flagship with 1M token context
- ✓Native agentic tool calling: computer use, search grounding, multi-agent orchestration
- ✓Fully OpenAI-compatible API, zero SDK migration cost
- ✓Transparent pricing: $1.25/1M input, $4.25/1M output, cached input $0.15/1M
- ✓Free tier: 60 RPM / 2M TPM, enough for prototyping
- ✓Multimodal input: text + image + video + PDF, text output
⚠️ Cons
- ×Currently only 1 API model (Muse Spark 1.1), limited selection
- ×Public preview limited to US developers
- ×China access requires stable proxy
- ×No public affiliate program
- ×Vs Claude Sonnet 5 at $3/$15, Meta's $1.25/$4.25 is competitive on output, closer on input
🎯 Best For
Agentic tool-calling workflows; multi-step code generation and refactoring; multimodal understanding (image/video/PDF); enterprises needing Meta's official compatible API
💰 Pricing & Plans
| Model | Input ($/M tokens) | Output ($/M tokens) | Best For |
|---|---|---|---|
| Muse Spark 1.1 | $1.25 | $4.25 | Agentic tool-calling, computer use |
| Cached input | $0.15 | — | Repeated system prompts (10× cheaper) |
| Web search | — | $2.50/1k queries | Search-grounded answers (per-query fee) |
🔧 API & Developer Experience
- •API Style: OpenAI-compatible /v1/chat/completions — drop-in replacement by swapping base_url and api_key. No SDK migration needed for existing OpenAI codebases.
- •Base URL: https://api.meta.ai/v1 (public preview) — Meta's official endpoint for the developer platform.
- •Context Window: 1M tokens — among the largest context windows in any frontier-tier API, suitable for full-codebase analysis and long document processing.
- •Multimodal Input: Native support for text, image, video, and PDF inputs in a single request; text output only. Useful for OCR, video understanding, and document analysis.
- •Function Calling & Tools: First-party native tool calling with structured JSON-schema validation. Includes computer use, search grounding, and multi-agent orchestration primitives — among the most capable agentic tool surfaces available.
- •Streaming: SSE token streaming included by default; intermediate reasoning/tool traces are inspectable in stream chunks for agentic debugging.
- •Free Tier: 60 RPM and 2M TPM with free credits on signup for US developers — enough to prototype agentic workflows before paid usage.
🧠 Native Agentic Tool Surface
What sets Meta's API apart from other OpenAI-compatible providers is the depth of its first-party agentic tool surface. While most compatible APIs expose Function Calling as a thin schema-validation wrapper, Meta ships computer use (browser/desktop control), search grounding (real-time web search with citations), and multi-agent orchestration primitives as built-in tools — not third-party add-ons. For teams building coding agents, research agents, or browser-automation products, this means the agent's tool loop can stay within Meta's API surface rather than stitching together multiple vendors. The 1M token context window combined with computer use and search grounding is a particularly strong combination for long-horizon agent tasks that need to read, search, and act across many web pages or documents in a single session.
🌐 Regional Availability & Latency
Meta's API is currently in public preview and is limited to US developers. Access from mainland China requires a stable proxy or overseas server routing. From a US-based client, latency to api.meta.ai is typically 100-200ms for first-token streaming — comparable to OpenAI/Anthropic from the same region. For production workloads serving China-based users, an aggregator (such as OpenRouter with Meta routing, or domestic re-sellers) is the standard pattern. Direct connection from Chinese IPs is not currently supported at the network edge. The public preview also means model availability and pricing may shift before general availability — teams building on Meta's API should plan for API surface changes over the next 6-12 months.