On 2026-10-12 Meta introduced Muse Wonder and Muse Wonder Full, its most capable models yet for personal and professional agentic work. Official posts highlight Active Memory (carry and retrieve what matters across conversations) and Active Focus (persist through hard tasks, recover from errors, clarify and confirm before consequential actions). Muse Wonder is available today via the Meta Model API and Muse Code; Muse Wonder Full is available today for Muse Code subscribers. Third-party Artificial Analysis Intelligence Index (max reasoning) lists Wonder Full at 56 and Wonder at 52 (Spark 1.3 Max was 48). Update Oct 16: Muse Wonder is now in Cursor with 300k context (1M Max Mode at the same rates), $4/$20 per million tokens, and 46.1% on CursorBench per Cursor.
Key Takeaways
- ✓Announced 2026-10-12 via Meta Newsroom and @AIatMeta (official post)
- ✓Availability (Meta): Muse Wonder on Meta Model API + Muse Code; Muse Wonder Full for Muse Code subscribers only
- ✓Core mechanisms: Active Memory + Active Focus for long-horizon personal/professional agents
- ✓Third-party AA Intelligence Index (max): Wonder Full 56, Wonder 52 vs Spark 1.3 Max 48 (changelog)
- ✓Now in Cursor (2026-10-16):
muse-wonder-1.0, 300k / 1M Max Mode at same rates, $4/$20 per M tokens; Cursor reports 46.1% CursorBench at default effort (Cursor docs)
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
After rebuilding on Muse Spark, Meta still lacked a closed frontier pair for long-horizon personal and professional agents. Insurance claims, leases, finance/health work, and sustained coding need memory across sessions and persistence when tasks stall. Meta’s Newsroom post positions Muse Wonder / Wonder Full as its most capable models to date.
Architecture
The research blog highlights Active Memory (carry and retrieve what matters across conversations) and Active Focus (try alternatives, recover from errors, clarify, confirm before consequential actions). Evaluation mentions include Harvey LAB, FocusBench, and SWE-Marathon. Models were trained at the Prometheus gigawatt-scale data center on the Muse Spark scaling ladder.
Benchmarks
Vendor charts are on the research blog; a fuller safety report is said to be coming soon — we do not invent unpublished numbers. Third-party Artificial Analysis Changelog (12 Oct 2026) lists Intelligence Index (max): Muse Wonder Full 56, Muse Wonder 52, vs Muse Spark 1.3 Max 48.
Developer path
Muse Wonder is available via the Meta Model API (https://api.meta.ai/v1) and Muse Code; Muse Wonder Full stays Muse Code-only. Update 2026-10-16: Muse Wonder is now in Cursor (model docs, ID muse-wonder-1.0): 300k context by default, 1M in Max Mode at the same per-token rates, six effort levels (minimal to max, default high), priced $4 input / $0.20 cached read / $20 output per million tokens from prepaid credit. Cursor reports 46.1% on CursorBench at default effort, the highest of any Muse model. Teams/Enterprise admins must enable the Meta provider first.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.