Muse Spark 1.2 isn't the laptop-sized model Meta gave away last week — it's the one built to fight Claude Opus 5 and GPT-5.6 head-on, and Zuckerberg just said it's going free too.
Mark Zuckerberg posted a video to Instagram on August 10th and said the sentence Meta watchers have been waiting on for two years: the weights for Muse Spark 1.2 — Meta's actual flagship model, the one that powers its coding agent and trades blows with Anthropic and OpenAI's best — are coming out. Not eventually. Not maybe. 'Soon,' he said, is when anyone will be able to download and run it themselves.
This is not the model Meta gave away last week
If you read Meta's Muse Glimmer release last week and assumed Meta had already gone all-in on open weights, that's only half true. Glimmer is a 30B dense model Meta compressed down to under 20GB specifically so it would fit on a single RTX card or a MacBook — a local-use model aimed at hobbyists and agentic tinkering, and it already runs meaningfully close to a much pricier Nvidia card on a $1,299 AMD one. Muse Spark 1.2 is a different animal. It's the model Meta actually competes on the frontier leaderboard with — the one enterprise customers pay API fees for, benchmarked directly against Claude Opus 5 and GPT-5.6 Terra, not against Gemma or other lightweight open rivals.
5th
Current rank
Artificial Analysis Intelligence Index
3 closed frontier models
Benchmarked against
Opus 5, GPT-5.6 Terra, Gemini 3.6 Flash
Muse Code
Powers
Meta's terminal coding agent
April 2026
Proprietary since
launch after Llama retirement
That timeline matters more than the essay does. Meta killed the Llama brand and launched Muse Spark as a closed, API-gated model in April — a move that read at the time like Meta following Anthropic and OpenAI toward the paywall, a genuine surprise from the company that built its whole AI identity on 'open.' Four months later, it's reversing course on the flagship model specifically, not just shipping a scaled-down local sibling and calling it a win.
Muse Code, Meta's terminal coding agent, currently runs on the proprietary version of Spark 1.2. · Unsplash
Zuckerberg's real argument isn't really about developers
Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data. US policy must reduce this additional friction if we want American open source models to lead over time.
Mark Zuckerberg, August 2026 essay
That's the actual thesis of the 6,500-word essay Zuckerberg published alongside the announcement, and it's a policy argument dressed up as a product release. The target isn't really Anthropic or OpenAI — it's DeepSeek and Moonshot, the Chinese labs whose open-weight models have gotten uncomfortably close to the American frontier over the past year, for free, while US labs argue about training-data restrictions. Giving away Muse Spark 1.2 is Meta positioning itself as the American answer to that gap, and lobbying Washington to loosen the rules Zuckerberg says put US labs at a disadvantage.
What's still unconfirmed
An exact release date — 'coming weeks' is Zuckerberg's phrase, not a calendar entry
The license terms Spark 1.2 will actually ship under
Whether the released version matches the API model's full context window and tool-use, or ships as a distilled variant
Hardware requirements — Spark 1.2 is frontier-scale, so expect multi-GPU territory, not the single-card story Glimmer told
Is Muse Spark 1.2 the same as Muse Glimmer?
No. Glimmer is a separate 30-billion-parameter model built to run on one consumer GPU or a laptop. Spark 1.2 is Meta's flagship — the model that powers Muse Code and benchmarks against Anthropic's Opus 5 and OpenAI's GPT-5.6 Terra.
When can I download Muse Spark 1.2?
Meta hasn't given a date. Zuckerberg said the weights are coming 'in the coming weeks' as of August 10, 2026.
Will Muse Spark 1.2 run on consumer hardware?
Almost certainly not on a single GPU the way Glimmer does. It's a frontier-scale model competing with Opus 5 and GPT-5.6 Terra — expect it to need serious multi-GPU or datacenter-class hardware to run at full precision.
Why is Meta doing this after going proprietary in April?
Zuckerberg's essay frames it as a competitive response to Chinese open-weight labs like DeepSeek and Moonshot, paired with a push for US policy that gives American labs more freedom to train on data. It's as much a Washington lobbying move as a product release.
What I'd actually watch for
The number that will tell you whether this matters is the license, not the benchmark score. If Meta ships Spark 1.2 under real Apache- or MIT-style terms, this is the biggest open-weight release of the year and it resets what 'open' means at the frontier. If it ships under another Llama-style community license with usage caps, it's a headline dressed as a gift. I'd also watch how Qwen3-Coder and other open coding models respond — a genuinely open Spark 1.2 with Muse Code attached would be the most serious local alternative to GitHub Copilot anyone has shipped, in the same way gpt-oss reset the calculus for local coding models when OpenAI went open earlier this year.
