Meta releases Muse Code in beta, a terminal coding agent powered by Muse Spark 1.2, a coding-focused model priced at $1.25/1M input and $4.25/1M output tokens

Meta is rolling out its first coding agent called Muse Code as the company tries to challenge leading AI labs Anthropic and OpenAI.

CNBC 2026-08-05 Jonathan Vanian

Context & Ripple Effects

Muse Code follows Meta’s July expansion of Muse Spark 1.1 into a public API preview with stronger coding capabilities, turning the model line from a developer API into a terminal-based coding-agent product. Meta is retaining the same listed standard token rates previously associated with Muse Spark 1.1.

The beta arrives alongside a lower-cost contributor tier for Muse Spark 1.2 that trades prompt use for training access. Together, the releases give Meta two routes into developer workflows: a packaged agent and a lower-cost model-access option.

First-order effects

Developers can now use Meta’s first terminal-based coding agent in beta, with Muse Spark 1.2 serving as the underlying coding model at the stated standard API prices.
Meta moves from positioning Muse Spark as a model for coding to owning an agent interface, placing Muse Code directly against the coding-agent offerings of Anthropic and OpenAI.

Second-order effects

Coding-agent buyers can compare Meta’s terminal workflow and listed token pricing with rival offerings, making both model cost and agent usability more visible purchase criteria.
The contributor tier creates a lower-priced access path for developers willing to let Meta use prompts for training, separating price-sensitive experimentation from standard API usage.

Third-order effects

If Meta continues pairing Muse models with task-specific interfaces, competition shifts from standalone model APIs toward control of the developer workflows where agents are deployed.
Meta’s planned use of agents across its products points to a broader embedded-agent strategy in which model distribution and product integration matter alongside coding-model performance.

The trend: AI labs are moving from selling coding-capable models to packaging them as agents while segmenting access by price and data-use terms.

Discussion

@teortaxestex @teortaxestex on x
LMAO Muse Spark has cheaper input, output *and, uniquely, cache hits* than V4-Flash. They don't care about margins, they want the data
@synthwavedd Leo on x
If you use their “contributor” slug - i.e. opt in to data collection for Meta to “improve their products” - you can use Spark 1.2 at ridiculous pricing of $0.10/Mtok input and $0.20/Mtok output [image]
@stocksavvyshay Shay Boloor on x
$META launches Muse Code in beta which is powered by Muse Spark 1.2 to run multiple coding agents across one project without conflicts. Most of the benchmark gain comes from the software around the model showing the product now matters as much as the AI itself. [image]
@deryatr_ Derya Unutmaz on x
Muse Code powered by Muse Spark 1.2 from Meta looks like a very interesting terminal coding agent! Worth trying out.
@mweinbach Max Weinbach on x
So I did have this over the weekend to try out a bit and uh The harness is good, the model isn't great Def better than the previous Muse Spark 1.1, but I wasn't a huge fan. It was an early build of the harness, so maybe that's it.
@altryne Alex Volkov on x
Everyone wants a harness (and data to train on) to be developed with their models. Everyone now understands that this is the most likely path to a generalized intelligence. Anthropic with Claude Code saw it first, OpenAI very quickly and very impressively caught up and
@scaling01 @scaling01 on x
RIP GOOGLE LONG LIVE META
@finkd Mark Zuckerberg on x
Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update. [image]
@cheatyyyy @cheatyyyy on x
Jesus Christ you should use Muse Spark 1.2 as your default cheap model it is so SO fast, I'm averaging 200 TPS in general use, not going to use anything but xhigh for early impressions but man it is so snappy and responsive in pi Meta's compute shining here as expected, excited […
@craigweiss Craig Weiss on x
wow, muse code is actually in the same league as codex and claude
@shengjia_zhao Shengjia Zhao on x
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
@claireszhou Claire Zhou on x
We are live with MSL's first coding agent!! This has been the most exciting work — tying Muse Spark coding capabilities to a product that developers across the world can use, and we're just getting started 🚀
@chengxuzhuang Chengxu Zhuang on x
The team spent days and nights (and nights and nights...) on cooking this model. We made big improvement on coding! Please test the model out and let us know what you think!
@mattdeitke Matt Deitke on x
We're releasing Muse Code and Muse Spark 1.2 today! It's a fast update from Muse Spark 1.1 a few weeks ago with more focus on coding and a new harness. Excited for you all to use it :) More evals below 👇 [video]
@pratanchandani Prashant Ratanchandani on x
The team has been cooking Muse Code, and we're excited to launch it in beta now. We're bringing with it Muse Spark 1.2 - a coding focused update, still with our emphasis on making sure we build models which provide leading cost-competitive intelligence to everyone. Sign up
@alexandr_wang Alexandr Wang on x
muse code in beta is here: our first coding agent powered by our latest model, muse spark 1.2. one command to install and start building. get it through Meta Model API.
@kylebrussell Kyle Russell on x
like deepseek v4 flash 0731 this would basically feel free, sipping pennies
@giffmana Lucas Beyer on x
I think this is a very nice deal for random hobby stuff. In my experience, the model is an absolute beast in terms of execution. It's not quite as clever/clean as Sol/Fable in complex situations yet, but it easily holds up to Luna/Terra/Opus while being super fast. (At least in
@scaling01 @scaling01 on x
they actually give you the model almost for free if you allow them to train on your data [image]
@kimmonismus @kimmonismus on x
Really to see Meta catching up and continuing to release new titles quickly. Muse Spark 1.2 is a very good terminal agent that holds its own very well. The only caveat: it's being compared to Terra 5.6 xhigh instead of Sol high. [image]
@haider1 Haider on x
how quickly the tables have turned Meta is starting to look like Google—moving fast, shipping AI products, and finally building momentum meanwhile, mighty Google suddenly looks like last year's Meta: confused, reactive, and somehow watching everyone else move faster
@chris_j_paxton Chris Paxton on x
including a bash command in your announcement tweet is a pretty great move
@tianheyu @tianheyu on x
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
@heaney555 David Heaney on x
Wow. Meta is providing a GPT Terra & Claude Sonnet class AI coding agent for 20x lower cost to developers if they'll allow the output to be used to train future models. That's a highly disruptive, though Faustian, offer.
@xeophon Florian Brand on x
@scaling01 Actually insane deal that might be very much worth it
@theo @theo on x
It has literally no awareness of Muse at all lmfao [image]
@theo @theo on x
Bad news. Muse is so gemini-flavored that it literally thought Muse was a codename for Antigravity... [image]
@quxiaoyin Xiaoyin Qu on x
Glad to see my ex-employer up their game. We need more competition.
@artificialanlys @artificialanlys on x
Meta has released Muse Spark 1.2. It's their third release in four months and scores 54 on the Artificial Analysis Intelligence Index, significantly improving agentic knowledge work capabilities over prior releases and putting Meta next to SpaceXAI in a tie for third place among…
@finkd Mark Zuckerberg on x
@luckeyfaraday I'll have more to share on that soon.
@cryptopunk7213 @cryptopunk7213 on x
i never thought i'd see a meta ai model reach parity with fable and gpt 5.6 at coding but here we are muse spark 1.2 beats gemini 3.6 flash and grok 4.5 in deepSWE. this is metas second model release in <4 weeks. they truly pulled off a miracle and the single common denominator […
@openrouter @openrouter on x
Muse Spark 1.2 from @AIatMeta is live on OpenRouter alongside expanded global access to both Muse Spark models. At $1.25/M in and $4.25/M out, the model builds its position as one of the most price-efficient, high-intelligence models on OpenRouter. [image]
@giffmana Lucas Beyer on x
While we can discuss and gossip how today is the end of an era for one of the BigCos... today may also be the beginning of an era for another of the BigCos: time to give the Muse Code beta a shot!
@luckeyfaraday Luckey Faraday on x
@finkd Will Muse Code be open source?

Chronicles