Put your ideas to work with Muse Wonder and Muse Wonder Full
Today we’re excited to launch Muse Wonder on Muse Code and Meta Model API at $4 input / $20 output per million tokens (USD).
Muse Wonder Full, available exclusively by subscription in Muse Code, gives developers a frontier model for their most complex workflows. Developers can start building with Muse Wonder Full on both the Power Usage plan and new Maximum Usage plan, which brings an 80x limit over the Everyday Usage plan and the same five-hour and weekly resets.
Install Muse Code today to start building:
For more details on the models and their research foundations, check out the Muse Wonder post on the AI at Meta blog.
TL;DR
- Muse Wonder - launching on Muse Code and Meta Model API at $4/$20 per million tokens, Muse Wonder brings comparable performance to Muse Wonder Full.
- Muse Wonder Full - available exclusively through the Power Usage and Maximum Usage Muse Code subscription plans, Muse Wonder Full is our most advanced model yet, bringing SOTA coding performance, long-running agents, and advanced computer-use capabilities to tackle developers’ most critical workflows.
- Muse Code Maximum Usage plan - a new subscription plan in Muse Code that includes 80x the Everyday Usage limit and access to Muse Wonder Full (also available in the Power Usage plan).
Capabilities at the frontier
We built the Muse Wonder series to tackle the complex coding and agentic work that developers face every day:
- Multi-step reasoning - chains reasoning over long horizons, sustaining coherent plans across complex, extended workflows.
- End-to-end agentic task completion - multi-step planning, sequential tool invocation, failure recovery, and long-horizon task execution, working within agent scaffolds to write and debug code and resolve multi-turn requests from start to finish.
- Coding agents - streamline complex engineering workflows like writing and resolving across large repos, including multi-file refactors, debugging, and long-running tasks.
- Tool use - tool invocation across extended multi-turn workflows, diagnosing errors and recovering from tool-call failures.
- Multimodal reasoning - interpreting screenshots, charts, documents, images, and video alongside conversation. Audio-dependent use cases require confirmation of audio support on the release surface.
Long-running agents need the right context at each step, so we built the Muse Wonder series with active memory, which carries important context across conversations, retrieves the right details for the task at hand, and keeps goals, constraints and earlier decisions available as work progresses.
Active focus helps agents keep working when a task gets difficult. The Muse Wonder series is trained to try alternative approaches, work through blockers and follow through to completion. It is also trained to ask for clarification when needed and check in before important actions, so developers can steer the model to better understand their intent and accomplish what they truly want.
Muse Wonder Full tops coding benchmark leaderboards, including DeepSWE and SWE-Marathon in independent third-party runs, outperforming every other model evaluated, and competes with the best across agentic evals:
Category
Benchmark
Muse Wonder Full (max)
Muse Wonder (max)
Claude Fable 5.1 (max)
GPT-6 Astra (max)
Claude Opus 5.5 (max)
Personal and professional agents
Coding
General
59.1
13.1
59.1
62.3
45.4
33.2
75.3
85.5
57.1
79.0
59.0
56
54.3
10.0
56.6
60.5
39.1
30.8
70.5
85.6
52.0
78.8
53.1
50.0
52
49.9
6.4
58.9
62.1
32.1
27.6
73.1
79.0
52.0
67.4
45.6
55.5
53
54.5
8.6
53.5
65.3
41.6
34.2
66.1
81.4
59.1
74.1
42.5
52.5
53
55.2
4.2
58.6
65.6
41.2
30.6
75.4
81.5
59.6
74.2
46.3
70.5
58
For more details about our evaluations, see our report.
Put your ideas to work
Muse Wonder
Frontier coding, long-running agents, and advanced computer use. Plans long tasks, delegates to sub-agents, and recovers from errors.
Muse Wonder Full
Our most advanced model yet, bringing SOTA performance to tackle developers’ most critical workflows.
Access
Muse Code subscription only
Included with a Muse Code subscription
Access on the Power Usage and Maximum Usage plans.
Already have an existing project built around the OpenAI SDK? You’re just one base_url change away from generating your first API call with Muse Wonder. Generate an API key at dev.meta.ai, point your existing OpenAI-compatible client at Model API, keep your code, and set the model to muse-wonder-1.0.
If this is your first time building with Muse models, the Meta Models cookbook walks you through the basics and best practices, so you can get started quickly.
Whether you’re starting a new build with Muse Wonder or just getting started with Meta Model API, we can’t wait to see what you build.
Build with Muse Wonder · Read the docs · Get started with Muse Code