Meta releases Muse Code agent and Muse Spark 1.2 model for developers
Meta launches a terminal coding agent and upgraded AI model to compete with OpenAI and Anthropic, offering cheaper pricing for data-sharing tier.
Conversation activity · last 5 days peak 35/hr
Summary, timeline and people extracted by Claude from 57 items across 9 sources · 17h ago. Quotes are verbatim.
Meta released Muse Code, a beta terminal coding agent powered by its new Muse Spark 1.2 model, designed to handle complex software engineering tasks across large repositories with persistent subagents and long-horizon planning. The company also introduced a "contributor" pricing tier at $0.10 per million input tokens and $0.20 per million output tokens for users willing to let Meta train on their prompts, representing a 10-20x discount versus standard pricing. The release drew mixed reactions, with observers questioning benchmark comparisons and Meta's competitive positioning versus Chinese labs and frontier models.
- Meta launched Muse Code, a terminal coding agent with persistent async background subagents for complex software engineering tasks, powered by Muse Spark 1.2.
- A "contributor" pricing tier offers 10-20x discounts ($0.10-$0.20 per million tokens) for users consenting to data training, comparable to DeepSeek V4 Flash pricing.
- Meta retroactively added data usage terms to existing free credits, stating content may be used for product improvement without prior notice.
- Observers questioned Meta's competitive claims and benchmark methodology, noting losses to Anthropic's Opus and comparisons against lower-tier OpenAI models rather than frontier alternatives.
How it unfolded
-
Reaction Open-source advocates call for model weights release
A commenter urges Meta to publish model weights, arguing the company's current position in rankings would support releasing open-weight versions without harm.
“Meta, please go back to publishing weights. Your models aren't top-tier, this wouldn't hurt you at all at your current position in the rankings.”
ersiees · Hacker News ↗ -
Reaction Security concerns raised over billing limits
A commenter expresses concern about using the service without hard spending limits, noting only email alerts are available for credit card protection.
“I entered my credit card, but can not set a limit. The best I can do is get an email alert. I feel like I am one oopsie away from getting a 100 dollar bill.”
sams99 · Hacker News ↗ -
Reaction HN community discusses pricing discrepancy
Hacker News commenters note the contributor tier represents a 10x input and 20x output token discount versus standard pricing ($1.25 and $4.25 per million tokens respectively), and compare it to DeepSeek V4 Flash pricing.
“Meta is offering a 10x discount on input ($0.10 vs. $1.25/Mtok) and 20x discount on output ($0.20 vs. $4.25/Mtok) if you opt in to let them train on your data.”
tristanj · Hacker News ↗ -
Analysis Critics question benchmark comparisons
A commenter raises concerns about Meta's marketing strategy, noting the company compared against OpenAI's mid-tier model Terra rather than Sol, and lost benchmarks against Anthropic's Opus on most tests.
“They chose to compare against Open AI's mid tier model Terra instead of Sol and still lost some benchmark against it. They left Opus in and got beat in all but one benchmark.”
WhitneyLand · Hacker News ↗ -
Report Meta introduces contributor pricing tier
Wall Street Journal reports Meta is offering Muse Spark 1.2 pricing tiers, including a cheaper "contributor" tier at $0.10/1M input and $0.20/1M output tokens in exchange for allowing user prompts to be used for training.
-
Reaction Data usage terms changed for free credits
A commenter notes that Meta added small print to free credits stating "your content may be used for product improvement," which was not present when Muse Spark 1.1 launched, creating retroactive data usage terms.
“While using free credits your content may be used for product improvement”
bradfa · Hacker News ↗ -
Event Meta releases Muse Code and Muse Spark 1.2
Meta announced Muse Code (beta), a terminal coding agent powered by Muse Spark 1.2, its new coding-focused model. The agent features async background subagents, replay-safe runtime design with event logs, and bundled skills like /plan, /grill, and /goal for complex software engineering tasks.
What people are saying verbatim
“Muse Code takes on complex software engineering tasks across large repositories: planning changes, writing code, and validating the results.”
Meta, Official announcement · Meta research blog
“This marks our next step toward the frontier, with larger and much more capable models on the way.”
Meta, Official announcement · Meta research blog ↗
“Meta is offering a 10x discount on input ($0.10 vs. $1.25/Mtok) and 20x discount on output ($0.20 vs. $4.25/Mtok) if you opt in to let them train on your data.”
tristanj, HN commenter · Hacker News ↗ · Aug 5, 4:34 PM
“While using free credits your content may be used for product improvement”
Meta, Small print in terms · Meta developer terms ↗
“They chose to compare against Open AI's mid tier model Terra instead of Sol and still lost some benchmark against it.”
WhitneyLand, HN commenter · Hacker News ↗ · Aug 5, 4:25 PM
“Meta, please go back to publishing weights. Your models aren't top-tier, this wouldn't hurt you at all at your current position in the rankings.”
ersiees, HN commenter · Hacker News ↗ · Aug 6, 9:14 AM
“Yet more evidence that the most important characteristic of any model these days is long-sequence agentic tool calling.”
Simon Willison, Technology commentator · Simon Willison blog ↗
Voices from the web unedited
-
If you got the $20 in free credits from Meta for signing up when muse-spark-1.1 was release, please note that there's now small print stating "While using free credits your content may be used for product improvement" which was not present at muse-spark-1.1 launch when the credits were given out.If you don't mind Meta retaining your data, the…
-
For a local agent to be practical, generation latency must be low enough to maintain workflow continuity. To run Muse Glimmer on consumer hardware without degrading quality, we used quantization to shrink the language model to under 20GB and a lightweight DFlash drafter model to [image]
-
Muse Shimmer 30B — very cool that they're launching it with speculative decoding instead of butchering the core model with MoE — research.meta.ai/blog/introdu... [image]
-
They chose to compare against Open AI’s mid tier model Terra instead of Sol and still lost some benchmark against it.They left Opus in and got beat in all but one benchmark.Nothing wrong with trying to improve, but why the marketing games?Instead of trying to say in the post you’re “closer” to frontier, first set a clear goal to beat the Chinese…
-
LFG: Meta's Muse Glimmer is a remarkably capable model for its size: 30B parameters, local deployment, and the best reported result on 12 of 24 benchmark rows against Gemma4-31B and Qwen3.6-27B. By my count, it beats Gemma on 19 of 24 rows and Qwen on 14. Its strongest area is [image]
-
Tibo took care of the reply: https://x.com/thsottiaux/status/2085229896968651198?s=20And someone corrected the Zucked graph:
-
Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows. Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on [image]
-
https://pbs.twimg.com/media/HO-59jQaoAA_JZ1?format=jpgVery interesting they have a way cheaper "contributor" version "used to improve our products", how much of that is price discrimination vs the data being that valuable?Roughly DeepSeek V4 Flash pricing, though you can get V4 from providers that don't train on your data
-
Build with Meta's new open-weight model on AMD. AMD is enabling day-zero support for Muse Glimmer, the new open-weight model from Meta Superintelligence Labs on AMD Ryzen™ AI Max+ systems and Radeon™ AI PRO R9700 GPUs. With @lmstudio making local AI accessible and Lemonade [image]
-
Stunning to see so many comments here very eager to try it out and cheering for it. Meta is not one bit better (and in reality worse) than Larry Elisson or Palantir but I doubt we'd be seeing this if it'd be Oracle Code or Palantir Code.Though I guess that there's such a big population of (ex-)Meta employees here that this might explain it.