Oct 19, 2026
Meta's Muse Wonder Joins the Frontier Tier as Gemini 4 Argon Tops LMArena
This Week in LLMs, Monday, October 19, 2026. Covers October 12–18.
TL;DR
- Meta released Muse Wonder, its first frontier-class models. Artificial Analysis scores the top version 56, up from 48 for Meta's previous model (Artificial Analysis).
- Google's Gemini 4 Argon opened to developers and subscribers on October 13 and now sits first on the LMArena text board, still marked preliminary (Google, Arena).
- DeepSeek shipped V4.1-Flash-1017 with a 1M-token context at $0.30/$1.20 per million tokens at peak, and the White House AI task force adopted its charter (DeepSeek, Semafor).
Top Story: Meta Moves Into the Frontier Tier
On October 12, Meta launched Muse Wonder and Muse Wonder Full, which it calls its most capable models for agentic work (Meta). Muse Wonder is available through the Meta Model API. The Full version is limited to Muse Code subscribers.
Artificial Analysis rates Muse Wonder at 52 and Muse Wonder Full at 56 on its Intelligence Index, against 48 for Muse Spark 1.3. The 56 puts Full level with several Claude Opus 5.5 and Sonnet 5.5 settings, just behind Opus 5.5 at 58 (Artificial Analysis).
Price rose with capability. Muse Wonder costs $4 input and $20 output per million tokens, up from $1.25 and $4.25 for Muse Spark 1.3. Artificial Analysis puts its cost at $4.52 per Index task, about 2.8 times the older model. Meta's claim that it beats rivals on its own memory benchmark is Meta's testing, not independent verification (TechCrunch).
Why it matters: Business users gain another frontier option, but should compare cost per finished task, not headline scores. Builders should measure input-token use per task before switching, since Artificial Analysis found it higher than Muse Spark 1.3.
Leaderboard Watch
| Model | Board | Old rank | New rank |
|---|---|---|---|
| Muse Wonder Full (max) | Artificial Analysis Intelligence Index | New entry | #4, score 56 (tied) |
| Muse Wonder (max) | Artificial Analysis Intelligence Index | New entry | #10, score 52 |
| Gemini 4 Argon (High) | LMArena text | New entry | #1, 1524 (preliminary) |
| Muse Wonder (Max) | LMArena text | New entry | #2, 1509 |
Sources: Artificial Analysis, Arena. These are new entries, so there is no prior rank. The Arena page dates its list October 15 and flags Argon as preliminary with about 4,900 votes, so expect movement.
Model & Pricing Changes
- Gemini 4 Argon: $2 input and $10 output per million tokens through November 30, then $4 and $20. It uses a shared 1M-token budget for input and output (Google).
- DeepSeek V4.1-Flash-1017: 1M context, 384K maximum output, $0.30/$1.20 at peak and half that off-peak (DeepSeek pricing). Artificial Analysis scores it 45 (Artificial Analysis).
- Muse Spark Open: Meta's open-weights model, listed at 816B parameters with a 1M context and a restricted commercial license. It scores 39 (Artificial Analysis).
- NVIDIA Nemotron 3.5: Super VL scores about 26 and Bolt 74B-A9B about 21, both open weights (Artificial Analysis).
- Reflection Beam: a first model with open weights, 501B parameters, a 1M context and a score of 30 (Artificial Analysis).
Policy Watch
- White House task force: It adopted its charter after meeting October 15, starting a 120-day window. It commissioned an advisory-council report that includes model control, and is drafting incident-reporting protocols for labs. Enforcement is unresolved (Semafor).
- China export draft: On October 16, two ministries proposed adding 40 technologies to the export-control catalogue, 23 of them AI-related. Comments close November 15. Technology already in the public domain is excluded (MOFCOM).
Tools Worth a Look
- DeepSeek Search & Fetch API: open to all developers, returning ranked web results or a page as Markdown (DeepSeek).
- FrontierCode on Artificial Analysis: a coding test that scores whether changes to real projects would be accepted by maintainers, not just whether tests pass. Claude Opus 5.5 leads at 52% (Artificial Analysis).
Events (Next 30 Days)
- Oct 20–21: AI & Big Data Expo Europe, Amsterdam (calendar)
- Oct 27–29: ODSC AI West, Burlingame (calendar)
- Nov 4–5: Future of AI Summit, London (calendar)
- Nov 10–12: Agentic AI Week USA, Atlanta (calendar)
- Nov 15: China export-control comment period closes (MOFCOM)
Keep Up With the Rankings
See how every model stacks up on the LLM1 leaderboard, and subscribe to the LLM1 newsletter for this roundup every Monday.