{"slug":"meta-s-muse-agentic-product-family","label":"Meta's Muse agentic product family","item_count":3,"day_count":3,"source_count":3,"first_seen":"2026-08-09T12:48:04+00:00","last_updated":"2026-08-14T05:05:00+00:00","via_scout":true,"generated_at":"2026-08-29T20:04:41.762844+00:00","sources":["hackernews_ai","infoq_ai_ml","search_agent_engineering_news"],"days":[{"date":"2026-08-09","items":[{"title":"Meta Launches Muse Code, Its First AI Coding Agent to Take on Claude and Codex - Memeburn","url":"https://news.google.com/rss/articles/CBMie0FVX3lxTE9qMDYtbG9BZzRiaVdmMTZNR0ZPMzFMdVh0MkR3Y0N3bkNnS0xwYlVwUVpCUmstb3JLZGtDTEVzSnlQZlRpeThnbk9UQmpIQkxHQlVzTDhPeTRyN0xMRlBDcG1qanNEU2lvbDBHRDNfMVlrMVdwcmFQSERIUQ?oc=5","source":"search_agent_engineering_news","type":"news","summary_1line":"Meta Launches Muse Code, Its First AI Coding Agent to Take on Claude and Codex Memeburn","why_it_matters":"Matches feed focus: agent, codex.","sid":"31ab186f51e5e6dc","published":"2026-08-09T12:48:04+00:00","editor_note":"Establishes the competitive frame — Meta targeting Claude Code and Codex, not general chat."}]},{"date":"2026-08-10","items":[{"title":"Meta releases a 30B agent model that fits in ~17GB","url":"https://huggingface.co/meta-models","source":"hackernews_ai","type":"release","summary_1line":"Meta releases a 30B agent model that fits in ~17GB","why_it_matters":"Matches feed focus: agent.","sid":"a61dbc8a3d17f5c2","published":"2026-08-10T10:28:56+00:00","editor_note":"First hard sizing number, and the first evidence of a self-hostable Muse tier."}]},{"date":"2026-08-14","items":[{"title":"Meta Open-Sources Muse Glimmer: A 30B Local Agentic Model Optimised for On-Device Execution","url":"https://www.infoq.com/news/2026/08/meta-muse-glimmer/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","source":"infoq_ai_ml","type":"news","summary_1line":"Meta AI Research has introduced Muse Glimmer, a 30-billion-parameter open-weight model under the Apache 2.0 license, designed for local workflows. It enables autonomous agents and complex task execution on consumer GP...","why_it_matters":"Matches feed focus: agentic.","sid":"683e96e306332d1e","published":"2026-08-14T05:05:00+00:00","editor_note":"Names the model and confirms the Apache 2.0 license — the detail that makes it evaluable in production."}]}],"editorial":{"tldr":"Meta entered the coding-agent market on Aug 9 with Muse Code, pitched directly against Claude Code and Codex. A day later a 30B Meta agent model appeared on Hugging Face at roughly 17GB, the first sign that Muse was a product family with a local tier rather than a single hosted agent.","stale":false,"whats_new":"Meta open-sourced Muse Glimmer on Aug 14 — a 30B agentic model under Apache 2.0, sized to run autonomous agents and multi-step task execution on a consumer GPU.","why_it_matters":"An Apache 2.0 30B agent model that fits in ~17GB is the first Muse tier you can run in your own VPC or on a workstation — the option that matters where hosted coding-agent spend or code-egress rules rule out Claude Code and Codex.","take_for_builders":"Pull the weights and run your own tool-calling and SWE eval before treating Glimmer as a drop-in for a hosted coding agent — no independent benchmark against Claude Code or Codex has surfaced in this thread yet.","status":{"state":"Shipping · open weights","tone":"launch","changed":"2026-08-14","detail":"Muse is now two tiers: a hosted coding agent (Muse Code) and an Apache 2.0 local model (Muse Glimmer), with no published head-to-head numbers for either.","track":[{"label":"hosted agent only","detail":"Aug 9","tone":"launch","weight":30},{"label":"open weights added","detail":"Aug 10 → now","tone":"now","weight":70}]},"beats":[{"kicker":"LAUNCH","tone":"launch","headline":"Muse Code puts Meta in the coding-agent race against Claude Code and Codex","summary":"Meta's first AI coding agent, positioned head-on against the two incumbents platform teams already pay for.","sids":["31ab186f51e5e6dc"]},{"kicker":"THE TURN","tone":"turn","headline":"A 30B Meta agent model shows up on Hugging Face at ~17GB","summary":"The size is the signal: small enough for one consumer GPU, which makes Muse a family with a self-hosted tier rather than a hosted-only product.","sids":["a61dbc8a3d17f5c2"]},{"kicker":"NOW","tone":"now","headline":"Muse Glimmer lands under Apache 2.0 for local agent execution","summary":"Meta AI Research names the open-weight tier and licenses it permissively for autonomous agents on consumer GPUs.","sids":["683e96e306332d1e"]}],"open_questions":["Is the ~17GB Hugging Face listing the same checkpoint as Muse Glimmer, or a separate quantized build?","How does Glimmer score on agentic tool-calling and SWE benchmarks against Claude Code and Codex?","What context window and tool-calling schema does Glimmer expose to existing agent harnesses?","Does Muse Code run on the Glimmer weights, or on a separate hosted-only model?","What are Muse Code's pricing and code-retention terms for commercial use?"],"provenance":{"683e96e306332d1e":{"verified":2,"status_update":true}},"generated_at":"2026-08-28T20:11:03+00:00"}}