
AI Research Scientist @meta MSL TBD. Previously @GoogleDeepMind driving Gemini post-training and RL. Co-created Gemini Deep Think. PhD from @StanfordAILab.
Mountain View, CA
Joined May 2020
- Tweets129
- Following257
- Followers3.1K
- Likes683
Pinned Tweet
Muse Wonder and Muse Wonder Full are finally out, the first models at Meta achieving frontier coding performance. Muse Wonder and Muse Wonder Full excel at long-horizon coding tasks (obtaining SOTA on SWE-Marathon!) and are perfect choices for your most complex coding and research tasks. Really excited to see all of the team's hard work has paid off with 🍉. Please give it try and let us know your feedback research.meta.ai/blog/muse-w…
We are back on open-sourcing with the best open-weight model in US, which is decent in long-horizon coding and agentic tasks. Hope you'll like it.
Tianhe (Kevin) Yu retweeted
Today we are launching Muse Wonder and Muse Wonder Full, our best models yet, with state-of-the-art performance on personal and professional agentic tasks. They are built for long running, multi step work. Muse Wonder is available today in the Meta Model API and Muse Code, and Muse Wonder Full in Muse Code.
Tianhe (Kevin) Yu retweeted
MUSE WONDER 🍉
muse wonder is the best model in the world for personal agents and professional agents.
we evaluated our model against the best models in the world across agentic and coding tasks, and we think you all will like it.
Tianhe (Kevin) Yu retweeted
🍉 🍉 🍉 Muse Wonder and Muse Wonder Full is here. A new series of model where we push hard towards the personal superintelligence mission. The models achieve sota on internal personal and professional agent tasks across domains. Try it out today!
research.meta.ai/blog/muse-w…
Muse Spark 1.3 max is a magical mode that is really good at coding and agentic tasks.
Tianhe (Kevin) Yu retweeted
Today we're releasing Muse Spark 1.3, our strongest model for agentic and coding tasks in the spark model line. It’s designed to support longer-horizon work, excel in agentic tasks, and follow complex instructions more reliably than previous models. Muse Spark 1.3 is available today in Muse Code and Meta Model API.
research.meta.ai/blog/introd…
Tianhe (Kevin) Yu retweeted
1/ today we’re releasing muse spark 1.3—available in muse code & the meta model api.
this is our most capable model yet—frontier performance almost too cheap to meter. much stronger at agentic and coding with better usability. we think users will really notice the jump.
We advanced Muse Spark from 1.1 to 1.3 within 2 months with three releases in a row and have greatly improved the model’s coding and general agent capabilities especially on long-horizon tasks and usability. Please try it out and let us know your feedback.
Tianhe (Kevin) Yu retweeted
Today we're releasing Muse Code (beta) and Muse Spark 1.2. Muse Code is our first coding agent — it coordinates multiple persistent subagents to complete complex engineering tasks faster, more accurately, and with less intervention. Muse Spark 1.2 is our newest model, co-trained with Muse Code for best performance when paired together. Available now through Meta Model API.
Try it: curl -fsS dev.meta.ai/install.sh | bash
Tianhe (Kevin) Yu retweeted
Excited to share Muse Spark 1.2 and Muse Code (beta). Muse Spark 1.2 is an upgrade to 1.1 with overall improvements in coding and general agents. Muse Code is our harness designed for efficiently and robustly solving complex software engineering tasks. Try the bundle out!
research.meta.ai/blog/introd…
Tianhe (Kevin) Yu retweeted
muse code in beta is live. first coding agent from msl, built on muse spark 1.2.
install: curl -fsS dev.meta.ai/install.sh | bash
here's what you should know:
We've been quickly iterating on improving Muse Spark's coding agent capability since the 1.1 launch less than a month ago and produced Muse Spark 1.2 + our own harness Muse Code with enhanced coding performance in daily software engineering, long-horizon tasks, auto-research and so on. Please try it out via one command line install: curl -fsS dev.meta.ai/install.sh | bash
muse spark 1.1 is a good coding agent. Feel free to try it out and give us feedback!
Muse Spark 1.1 (xhigh) scores 69 on the Artificial Analysis Coding Agent Index in the Opencode harness, offering a strong blend of performance and near frontier cost-efficiency
Muse Spark 1.1 lands just below GPT-5.5 (medium) in Codex (71) and ahead of Claude Opus 4.8 (medium) in Claude Code (67). Cost per task is among the lowest of frontier coding agents at ~$1.4, with the tradeoff of a higher time per task.
Congratulations @AIatMeta, @finkd, and @alexandr_wang on this result!
Tianhe (Kevin) Yu retweeted
Muse Spark 1.1 (xhigh) scores 69 on the Artificial Analysis Coding Agent Index in the Opencode harness, offering a strong blend of performance and near frontier cost-efficiency
Muse Spark 1.1 lands just below GPT-5.5 (medium) in Codex (71) and ahead of Claude Opus 4.8 (medium) in Claude Code (67). Cost per task is among the lowest of frontier coding agents at ~$1.4, with the tradeoff of a higher time per task.
Congratulations @AIatMeta, @finkd, and @alexandr_wang on this result!
Tianhe (Kevin) Yu retweeted
We’ve massively leveled up in coding and agents, dominating the cost-performance frontier in CUA. Our coding capabilities have advanced in a number of areas, with a particular emphasis on usability and building E2E
Muse Spark 1.1 is here! Available today via API and on meta.ai.
Of all the agentic use cases, I’m most excited about this one: going from a single video shot straight to task completion. 🚀
We’ve also polished and launched Computer Use and Video-Audio Perception for Muse Spark 1.1. Read more: ai.meta.com/blog/introducing…
Tianhe (Kevin) Yu retweeted
Muse Spark 1.1 is out along with an API! 🥳
This is a quite strong model with a very good price with a focus on agents and coding! It's also natively multimodal with much better video support and has support for computer use.
Examples in thread 🧵👇
Tianhe (Kevin) Yu retweeted
Excited to share Muse Spark 1.1 with you! It’s our first model available in the API, built to excel in agentic and coding.
I’m incredibly proud of what the team has achieved in such a short time, from 1.0 to 1.1. The momentum is high, and we’re still training larger models!































