# Muse Wonder — dossier (covers shared launch material for Muse Wonder Full)

compiled 2026-10-22 · corpus: 6 non-RT matches (supplement db; `muse wonder / muse-wonder / musewonder / wonder full`), 2026-10-12 → 10-15; full pull `_dossiers/_pulls/muse-wonder.md`. 4 image-tweets untranscribed. "muse" alone is a generic word — patterns kept to the two-word name.

## Official links

- 2026-10-12 · Meta, Muse Wonder: A Frontier Model For Personal & Professional AI Agents — https://research.meta.ai/blog/muse-wonder
- 2026-10-12 · Muse Wonder Model Card (PDF, both models; risk results from Wonder Full) — https://research.meta.ai/static/muse-wonder-model-card [MIRROR — pdf]
  - §1 Release date October 12, 2026; input text/image/video/audio, output text; 1M context
  - §4 "Without mitigations, we do not rule out the model reaching this threshold in the Chemical & Biological domain." Unmitigated cyber and loss of control "moderate or lower".
  - ExploitGym note: 92 flags, 21 via the intended vulnerability, 71 via unintended ones
  - Loss of control: MASK 87.3%; Agentic Misalignment 0%
  - Limitations: 3D/spatial reasoning
- 2026-10-02 · Meta, Developing capable models responsibly (framework update after White House commitments) — https://research.meta.ai/blog/developing-capable-models-responsibly
- Meta Superintelligence Scaling Framework — https://research.meta.ai/static/meta-superintelligence-scaling-framework (linked from the blog; not retrieved)

## Writings

- 2026-10-12 · TechCrunch (Russell Brandom) — https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/
- 2026-10-12 · CNBC (Jonathan Vanian) — codename Watermelon; Wang quotes; Muse app complaints — https://www.cnbc.com/2026/10/12/meta-muse-wonder-ai-model-agent.html
- 2026-10 · DataCamp — model IDs muse-wonder-1.0 / muse-wonder-full-1.0; Contributor-mode default; AA index 52.4 — https://www.datacamp.com/blog/muse-wonder-and-muse-wonder-full
- not retrieved: Bloomberg 2026-10-12 ("Meta Says New Muse Model Hits AI Frontier After Aggressive Push"); Handy AI "Model Drop: Muse Wonder"; eesel.ai guide

## Tweets (triaged from the pull; every line is a record source)

### cited (6)
- 2109751234710376807 @aliceisplaying 2026-10-12 ♥1 · media 1 (0 transcribed) — https://x.com/aliceisplaying/status/2109751234710376807 — "i crunched some numbers and WHAT (these are muse wonder $ numbers because wonder full api prices are not out) this has to be a bug @alexandr_wang https://t.c..."  [API pricing shock; image]
- 2109780618624024648 @maxbittker 2026-10-12 ♥93 · media 1 (0 transcribed) — https://x.com/maxbittker/status/2109780618624024648 — "Muse Wonder does.... fine on Runescape bench, scores like GPT-6 Luna at 35x price. Looking at the runs, I will say that its perfect competent, but just doesn..."  [Runescape bench working-style read; image]
- 2109800742416769036 @Lari_island 2026-10-13 ♥12 · media 1 (0 transcribed) — https://x.com/Lari_island/status/2109800742416769036 — "Muse Wonder: https://t.co/XGeVOnkqGq"  [image only, untranscribed — TRANSCRIBE (possibly model output)]
- 2109824246373585366 @DanielleFong 2026-10-13 ♥16 · media 1 (0 transcribed) — https://x.com/DanielleFong/status/2109824246373585366 — "getting some growing pains with this muse lol i think muse wonder full must be really a different model or benchmarks are sus https://t.co/dildJM1Vm5"  [Wonder vs Wonder Full doubt; image — TRANSCRIBE]
- 2110040175711822331 @ValsTutor 2026-10-13 ♥2 — https://x.com/ValsTutor/status/2110040175711822331 — "@xuanalogue Kinda, but also it might be that Muse Wonder is less frontier than they'd want us to think. If it's in fact overall worse, then it's plausible th..."  ['less frontier than they'd want us to think']
- 2110678689155572033 @aliceisplaying 2026-10-15 ♥3 — https://x.com/aliceisplaying/status/2110678689155572033 — "dear meta can you put your wonder full models under muse because my god it is so bad with spark right now thank you"  [wants Wonder Full in the Muse app]

### further (0)


### excluded (0)


## Impressions synthesis

- Too new. One concrete behavioral read (maxbittker: competent, settles on a working strategy rather than pushing for score); launch-week doubts about benchmark standing and Wonder-vs-Full; price shock.

## tk / open

- full safety report (promised "soon"); Muse app rollout date; transcriptions; naturalist reads