Evals, comparisons and research findings about Muse models.
li yin@panda_liyin · 10 Jul
muse spark 1.1 in AdaL Engineer beats Opus4.8 in Claude Code with 20% of the cost loop engineering, when done right, is beyond just running longer, its delivering better results when contexts are managed well and when workers are better prompted to stay honest. how GANs had https://t.co/dpNDBzffcL
thehype.@thehypedotnews · 10 Aug
muse spark 1.2 vs gpt 5.6 sol vs kimi k3 vs grok 4.5 – on two landing pages four coding agents built two desktop landing pages from scratch, then had to open them in a real browser, find their own bugs and fix them before they were allowed to hand anything over the setup: each https://t.co/elidTourH6
Artificial Analysis@ArtificialAnlys · 10 Aug
At 30B parameters, Muse Glimmer sits near the Intelligence vs Parameters frontier for open weights models: 5 points above Gemma 4 31B (Reasoning) at the same size, effectively matching Kimi K2.5 (Reasoning) at 33x fewer total parameters, and just behind Qwen3.6 27B (Reasoning), https://t.co/TKDNlTyTqy
John Yang@jyangballin · 9 Jul
My favorite demo from the launch: muse spark 1.1 + opencode runs evaluation of *itself* + mini-SWE-agent (by @KLieret, @closji, urs truly) on DeepSWE! https://t.co/mKDXrhK3yT
Artificial Analysis@ArtificialAnlys · 10 Aug
Muse Glimmer's gaps against its class concentrate in agentic evaluations: 953 Elo on GDPval-AA v2 against 1141 for Qwen3.6 27B (Reasoning), 1141 for Gemini 3.5 Flash-Lite, and 1004 for Kimi K2.5 (Reasoning), with Terminal-Bench v2.1 (52%) also behind Qwen3.6 27B (61%). The https://t.co/YCqwRhMtsi
thehype.@thehypedotnews · 12 Aug
muse glimmer 30b vs nemotron 3.5 lightning – on agentic coding two 30b open-weight agent models, released three days apart, both sold as the model you leave running. same harness, same tools, same tasks, two of them: repair a broken landing page, then write breakout from a spec https://t.co/mtIe7KNJmA
Simon Smith@_simonsmith · 13 Oct
Very early impressions of Muse Wonder in the Muse app: 1. No regressions I notice. In particular, no personality or writing regressions. Feels like the same entity I was talking to with Spark. 2. Better UI sense. Had it polish an app made with Spark and it found a really elegant way of positioning a footer “Next” button that kept going off screen in a quiz (it floated the button so it always stays in the same place rather than getting pushed down). 3. Caught subtle bugs in that same app. Not major ones, but ones that required a close reading of the code and understanding of how it was supposed to work and actually worked. One thing that’s not clear right now is whether Wonder also powers background tasks like goal summaries and the news feed. My current feeling is yes to to the news feed and no to the goal summaries based on how they’re still written (have some of the really bad AI writing tells, like trying too hard to have a voice, and self-referentially calling attention to things that they only need to show rather than also tell).
khaled@eltokh7 · 9 Jul
Muse Spark 1.1 seems like a very good model. I tested it on Stata Benchmark and it ranks 4th (!) beating Opus 4.7/4.8, tied with GPT 5.5. I like this benchmark because it often catches seemingly good model performing poorly on OOD tasks https://t.co/XNBL0EPEo6
Yair@finebutwhy · 12 Oct
meta's muse wonder came out today, so i gave it and opus 5.5 the same job: build a playable 3d game of my kitchen, first person, open the fridge, grab ingredients, cook one dish. one index.html, three.js from a cdn. one prompt, one shot, nobody touched the keyboard claude code on opus 5.5 at xhigh, muse code on wonder at xhigh (the api wonder; wonder full needs the bigger muse plan) opus 5.5: 11m40s, $2.10, ran first try: yes wonder: 3m42s, about $1, ran first try: yes first vibes for opus 5.5 xhigh vs meta wonder xhigh? what would you like to see tested next https://research.meta.ai/blog/muse-wonder
Louis-François Bouchard 🎥🤖@Whats_AI · 3 Sept
We just measured Meta Muse Spark 1.3 on our internal writing benchmark hoping for new SOTA. Unfortunately, it isn't... It enters at #24 of the 87 models we track, up from #31 for Muse Spark 1.1. As @alexandr_wang highlighted, it beats every Gemini configuration we have, https://t.co/90kdTpWE9P
adam roman@adamisnotroman · 13 Oct
I trusted Zuck so you don't have to. Muse Wonder used for a security audit for a small side project. About 4 small commits ended up costing me $7.50 and there was virtually no difference from the other frontier models. Maybe the commoditization of frontier models is happening sooner than we thought.
Peter James@heypeterjames · 21 Sept
By chatting normally with the Muse iOS app I was able to have it send a zip of its entire filesystem from the linux root to my Google Drive via the google connector. /home/hatch is the agents home and contains top level files live soul.md, identity.md, memory, tools, and more https://t.co/iAph5xp48Y
Fahd Mirza@fahdmirza · 13 Oct
💥 Meta just dropped Muse Wonder ♠ and I put it through 4 real tests with Hermes Agent 🔹Built for personal and professional agents, with Active Memory and Active Focus 🔹Meta's own charts put it ahead of Claude Opus 5.5 and GPT-6 Astra on DeepSWE and SWE-Marathon 🔹Test 1: build a 3D "floor is lava" game from scratch with Blender and Godot 🔹Test 2: a Brazilian churrasco picanha skewer animation in one HTML file 🔹Test 3: find the hidden bug in a nasty Oracle 8i query 🔹Test 4: read a Swahili headline from an image and translate it into 80+ languages 🔥Watch the full video below:👇 https://youtu.be/Be0fUeLhYYY?si=BXnbQcPProOJ7ju5 @alexandr_wang @AIatMeta
61 public posts describe it. Evals, comparisons and research findings about Muse models. Each card above quotes the poster or condenses their post, and links to the original.
No. Outcomes are what each poster reported. Built with Muse did not verify the savings, bookings or completions. 61 of the 61 posts are in the poster's own words; the rest are condensed summaries with a link back.
0 of these posts come with a prompt you can copy and adapt. Open Try the prompt on a card, paste it into Muse, and replace the bracketed parts with your own details.
Muse is invite only in most places. Built with Muse runs a community pool of invite codes: take one, redeem it in the Muse app, then add your own for the next person.
Outcomes are what the poster reported. Built with Muse did not verify the savings, bookings, or completions. Posts with text_source tweet are the author's own words. Posts with text_source field-guide-summary are condensed from the public post by the EveryAI Muse field guide (snapshot 2026-09-14) and link back to the original. Some entries were first catalogued by shipwithmuse.live. Data updated 2026-10-09.