Muse Wonder

MetaMCP Tool CallReasoning

meta/muse-wonder-1.0Closed API

Muse Wonder is a frontier multimodal reasoning model from Meta for personal and professional agents, long-horizon coding, and computer use. It combines Active Memory, which carries relevant context across conversations, with Active Focus, which keeps it working through blockers toward a goal.

Benchmark Pricing$4.00 / 1M in$20.00 / 1M out$0.20 cache read

Context: 1,048,576 tokens•Max Output: 131,072 tokens•Released: 2026-10-12

TECHNICAL SPECIFICATIONS

Context Window1,048,5761M tokens

Max Output131,072131.1K tokens

LicenseProprietaryCloud API only

Input Modalities

TextVisionPDFAudioVideo

Function / Tool CallingSupported (MCP Ready)

Structured OutputsGuaranteed Schema

MODEL CONTEXT PROTOCOL (MCP) READINESS

✓ Muse Wonder natively supports OpenAI / Anthropic format tool calling and is fully compatible with MCP servers and agent loops.

MCP Context Window Recommendations

With a context capacity of 1,048,576 tokens, this model can handle extremely large MCP resource manifests, complete database schemas, and multiple active server tool definitions simultaneously without degradation.

STANDARDIZED BENCHMARKS & EVALS

Harvey Legal Agent Benchmarkall-pass rate

Finance Agentaccuracy

56.6

HealthBench Professionallength-normalized weighted rubric score

60.5

AutomationBenchstrict pass@1

39.1

GDP.pdfscore

30.8

JobBenchmean reward

70.5

OSWorldmean partial score

72.6

OSWorldbinary completion rate

34.3

Terminal-Benchaccuracy

DeepSWEpass@1

78.8

SWE-Marathonpass@1

53.1

ProgramBenchshare of rollouts passing at least 95% of tests

Artificial Analysis Intelligence Indexindex score

API CALL INVOCATION EXAMPLE

// Example: Calling Muse Wonder via OpenAI SDK
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.META_API_KEY,
});

const response = await client.chat.completions.create({
  model: "meta/muse-wonder-1.0",
  messages: [{ role: "user", content: "Hello! Can you inspect MCP tools?" }],
});

console.log(response.choices[0].message.content);