--- title: "Introducing Muse Code and Muse Spark 1.2" url: https://daily.dev/posts/introducing-muse-code-and-muse-spark-1-2-vkdduwlvg source_url: https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2 type: article source: "Hacker News" published: 2026-08-05T19:47:58.338Z updated: 2026-08-06T03:01:40.284Z tags: ["llm"] reading_time: 4 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Introducing Muse Code and Muse Spark 1.2 **[Hacker News](https://daily.dev/sources/hn)** · 4 min read · 0 upvotes · 0 comments ## Summary Meta Superintelligence Labs has released Muse Code (beta), a terminal-based coding agent powered by the new Muse Spark 1.2 model. Muse Code features async persistent background agents that reduce latency on multi-step tasks, a local event log for replay-exact and restart-safe execution, and bundled skills like /plan, /grill, and /goal. Muse Spark 1.2 was co-trained with Muse Code, with significantly scaled compute on coding tasks, long-horizon training covering whole-repository generation and auto-research, and a self-improvement loop using Muse Spark 1.1 to generate training data. A case study demonstrates the agent iteratively optimizing GPU kernels (KDA and MLA) for NVIDIA Hopper GPUs over 1,000+ tool calls, achieving substantial improvements over baseline Triton implementations. Muse Spark 1.2 is available in Muse Code and via the Meta Model API. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Community take How the wider developer community reacted, aggregated from 1 discussion and 107 comments across hackernews (as of 2026-08-06). **TL;DR:** The community sees Muse Spark 1.2 as a solid but not frontier-level improvement, with the most excitement around the deeply discounted "contributor" pricing tier; skepticism centers on benchmark cherry-picking, data-privacy concerns with Meta, and whether Muse Code can displace Claude Code or Codex even internally. **Sentiment:** 25% positive · 45% mixed · 30% skeptical **The case for** - The contributor pricing ($0.10/$0.20 per Mtok) makes it a genuinely competitive alternative to DeepSeek V4 Flash for users comfortable sharing data. - Inference speed is notably high (~150–180 tok/s on OpenRouter), well above most DeepSeek providers. - The harness features — parallel agents in separate worktrees, crash recovery, and a local event log — are seen as genuinely interesting engineering. - The transparent two-tier pricing model (train vs. no-train) is praised as clear and honest compared to competitors. - The kernel optimization case study is highlighted as the most technically interesting part of the release. **The pushback** - Benchmark comparisons conspicuously omit the top-tier models (Sol, Fable) where Spark 1.2 would fare worse, undermining credibility. - Reports suggest Meta engineers themselves still prefer Claude Code or Codex, casting doubt on real-world quality. - The model is not open-weight, which contrasts unfavorably with DeepSeek's community releases. - Requiring a Facebook/Meta account creates friction for corporate users behind firewalls and developers who don't want personal accounts tied to work. - The rapid 1.1→1.2 release cadence (under a month) raises suspicion it may be a do-over prompted by Kimi K3 overshadowing the prior launch. - Some commenters distrust Meta's data-handling regardless of pricing tier, doubting the no-training guarantee is enforceable. **By community** - hackernews (mixed): Cautiously interested in the pricing innovation and harness features, but broadly skeptical of benchmark framing, Meta's trustworthiness with data, and whether the model is truly competitive with frontier alternatives. **Hottest debate:** Whether the discounted "contributor" pricing is a genuinely fair and transparent offer or whether Meta will train on all data regardless of which tier users pay for. **Open questions** - Will Meta release the model weights, as DeepSeek does? - Is Muse Code actually being adopted internally at Meta, or are engineers still using Claude Code/Codex? - Will a zero-data-retention endpoint be offered on OpenRouter or other third-party providers? - Does the contributor pricing guarantee hold legally, and how would a breach even be detected? - What is the actual parameter count of Spark 1.2, given the unusually high inference speed? **Highlights** > 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. https://developer.meta.com/ai/models/muse-spark/ > — [tristanj on hackernews · 8 comments](https://news.ycombinator.com/item?id=49188633) > So you think the only difference between the $1.25/million token plan and the $0.10/million token plan is that you pay them more to both lie to you and breach their contractual obligation to you? > — [simonw on hackernews](https://news.ycombinator.com/item?id=49191053) > Last I heard, everyone at Meta was using Claude Code. Any insiders know how Muse Code is doing internally? > — [wxw on hackernews · 3 comments](https://news.ycombinator.com/item?id=49188395) > Here's the Muse Spark 1.2 pelican: https://tools.simonwillison.net/markdown-svg-renderer#url=ht... I think it's a bit of an improvement on the Spark 1.1 pelican: https://simonwillison.net/2026/Jul/9/muse-spark-1-1/ > — [simonw on hackernews · 2 comments](https://news.ycombinator.com/item?id=49189380) > The most interesting thing here is the kernel optimization graph. It look like all models were still improving, when they cut off the experiment. It reminds me of a genetic algorithm. The graph is the same: long plateaus and then massive leaps. The only difference between the models seems to be how quickly they arrive. > — [andai on hackernews](https://news.ycombinator.com/item?id=49191106) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=49187575) · 177 points · 107 comments --- Tags: [#llm](https://daily.dev/tags/llm) [View this post on daily.dev](https://daily.dev/posts/introducing-muse-code-and-muse-spark-1-2-vkdduwlvg)