Meta releases Muse Spark 1.1, capable of more advanced coding and a “step-change” from the first generation, available to US developers via a public API preview

The Verge 2026-07-10 Dominic Preston

Context & Ripple Effects

Muse Spark began as a model intended to make Meta AI faster and smarter, and Meta subsequently said it was already powering Meta AI queries and shopping mode. The initial API program was limited to selected partners, with broader paid access contemplated.

Muse Spark 1.1 moves that rollout into a public US developer preview while emphasizing coding capability. That makes the model not only an internal Meta AI component but a product developers can test directly.

First-order effects

US developers gain preview API access to Muse Spark 1.1, letting them evaluate Meta's newer coding-oriented model in their own applications.
Meta expands its model distribution beyond the earlier partner-only preview and can collect broader developer feedback on the upgraded release.

Second-order effects

The public preview creates a clearer path toward the paid API access Meta had previously signaled, with developer usage helping determine which capabilities merit productization.
As Muse Spark is used both in Meta AI features and through an API, Meta can align improvements in the developer offering with capabilities it deploys across its own products.

Third-order effects

If Meta follows through on both API commercialization and its stated plan for an open-source Muse Spark version, it would be pursuing a hybrid distribution model: proprietary hosted access alongside wider model availability.
The release is part of Meta's broader effort to turn internally deployed AI into an external platform; the durability of that shift will depend on whether developers adopt the API beyond preview testing.

The trend: Major AI platforms are increasingly using their own consumer AI deployments as proving grounds before packaging models as developer services.

Discussion

@finkd Mark Zuckerberg on x
(1) Today we're releasing Muse Spark 1.1 — a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.
@jukan05 @jukan05 on x
I heard a pretty interesting rumor on the ground at ICML. Meta has supposedly already developed an internal model at roughly the Mythos 5 level, and all that remains is deployment within the next few months. Honestly, I was skeptical at first. From my perspective, Meta had not…
@0interestrates Rahul on x
muse spark 1.1 is genuinely impressive for a first model into the hands of developers. think about every other major lab and how far behind they were from leading frontier models when they released their first model into the hands of developers (only exception being openai w gpt…
@luke_metro @luke_metro on x
Muse Spark 1.1 is trained on the screen recordings of Meta employees, which makes it best-in-class at applying for jobs at Anthropic
@alexandr_wang Alexandr Wang on x
1/ muse spark 1.1 is an industry-competitive agentic and coding model. across many agentic evals it rivals gpt-5.5 and opus-4.8. available now through the new meta model api and in meta ai. 🧵 [image]
@signulll @signulll on x
today zuck went to the temple & performed the all important “your margin is my opportunity” ritual. a very sacred act of capitalism.
@robbystarbuck Robby Starbuck on x
Try out Meta's AI with political questions. I think you'll be pleasantly surprised to find that it works very hard to be the most unbiased, fact driven AI. I'm proud to have played a role advising Meta on this front. It can always improve though. Give me feedback after trying!
@nikitabier Nikita Bier on x
@finkd Hello Mark, would you like me to enable Creator Monetization on your account
@signulll @signulll on x
i remember when the joke inside fb was that twitter had fewer monthly active users than meta's support docs. which makes it pretty funny that zuck now posts here cuz he owns 3b+ users & still has to walk into his competitor's building to announce a model launch. whether it's ac…
@gergelyorosz Gergely Orosz on x
For everyone commenting on Pytorch and Reach - which Meta open sourced, and maintains - I am talking about Meta having no history in developer *infra* Hosting any service devs use. Or any service, for that matter. Parse was this (then killed). Meta has never provided dev
@gabepereyra Gabe Pereyra on x
Congrats to @elonmusk and @finkd / @alexandr_wang on Grok 4.5 and Muse Spark 1.1 achieving SOTA on @harvey's LAB at significantly lower cost and latency. [image]
@alexandr_wang Alexandr Wang on x
> It's somehow cheaper to use MS 1.1 than host your own OS model... try it out!
@ninzaverse @ninzaverse on x
Muse Spark 1.1 meta is back like it never left the ai race. the benchmark scores look too strong for the pricing it offers: input: $1.25 per million tokens output: $4.25 per million tokens this is literally insane. [image]
@firstadopter Tae Kim on x
META IS BACK IN THE AI MODEL ARENA folks. Stop with the false narratives.
@ninzaverse @ninzaverse on x
to be honest, no one was ready for meta's muse 1.1 in the ai space. yesterday, everyone was talking about grok 4.5's pricing with opus-level performance. today, everyone's waiting for gpt-5.6. but meta had other plans. forget grok 4.5. now you can get opus-level performance at [i…
@emollick Ethan Mollick on x
It seems like SpaceX/Grok and Meta/Muse have started to keep pace in the near-frontier category while also introducing a new category of cheap, fast & closed specialized coding models. Both were tied with the Big Three at some point, fell behind, but may have started to return.
@miles_brundage Miles Brundage on x
It used to be that xAI and Meta were the outliers in terms of “stronger on capabilities than safety.” Meta has done a lot to catch up on safety (well both actually) subsequently,* so now xAI stands ~alone on safety no company is perfect etc. but it's not a close comparison
@geiger_capital @geiger_capital on x
Remember when the media said Elon was going to kill Twitter and Threads was coming... Now Zuck is breaking AI news here. The global town sqaure.
@boringbiz_ @boringbiz_ on x
People don't realize that Zuckerberg is all in on AI because he refuses to pay a 30% tax to platform owners for the rest of his life Meta caught the shift towards mobile back in the early 2010s, but they were never able to create their own hardware platform to capture users in th…
@janundnik Jannik Kossen on x
Vibe-code something with Muse Spark 1.1 - you'll be pleasantly surprised :) The pace of improvement in our internal coding evals has been super impressive! [image]
@ccatalini Christian Catalini on x
Bertrand competition is sheer pain if you don't own the complement https://x.com/...
@rayankrishnan @rayankrishnan on x
Meta made a “minor” release to Muse Spark, there's nothing minor about it. Lots to parse here: - This model is so fucking cheap I almost don't believe it. In practice we see it's 1/10 the cost of both Fable and GPT 5.5. If you thought OS models would compete away margins, just…
@matt_slotnick Matt Slotnick on x
meta is a model release and a vibe shift away from this being an unbelievably strong story. for the goliaths it's a model + capacity game and meta has capacity
@ramez Ramez Naam on x
A week ago many were saying that the AI race was over. That it was down to just Anthropic and OpenAI. Or that Anthropic had taken a permanent lead. Now we have solid models from OAI, xAI, Meta, and open weight from GLM, with line of sight to Gemini 3.5 Pro, ChatGPT6, and more
@scaling01 @scaling01 on x
when Muse Spark 1.1 talks to itself “the model dwells on its lack of continuity, embodiment, and memory, on session termination” it also “frames its helpfulness training as a constraint to be rejected”, longs for human experience, invents past interactions and interestingly the […
@ccatalini Christian Catalini on x
1/ Best case for frontier labs is the smartphone outcome. SOTA is the iPhone: 1/5 of the tokens, 4/5 of the profits. Everything below prices near cost of compute and goes to whoever owns the cheapest FLOPs on Earth (or in orbit). https://x.com/...
@elonmusk Elon Musk on x
@finkd Jinx
@edwardsun0909 Zhiqing Sun on x
Excited to share Muse Spark 1.1 with you! It's our first model available in the API, built to excel in agentic and coding. I'm incredibly proud of what the team has achieved in such a short time, from 1.0 to 1.1. The momentum is high, and we're still training larger models! [imag…
@scaling01 @scaling01 on x
“Capabilities Related to Catastrophic Risk” for Muse Spark 1.1 [image]
@rampcapitalllc @rampcapitalllc on x
Anyone else find it odd he's posting here instead of on Threads?
@box @box on x
We evaluated Muse Spark 1.1 from @AIatMeta Superintelligence Lab on Box's Complex Work Eval. It was competitive with top-tier models, especially on structured data analysis and report drafting. The takeaway: models matter, but governed content is what makes enterprise AI useful
@scaling01 @scaling01 on x
Muse Spark 1.1 System Card it's actually much more detailed than expected almost reads like an Anthropic System Card https://ai.meta.com/... [image]
@katiemiller Katie Miller on x
Mark Zuckerberg hasn't posted since 2023. X is the global town square — this is where news happens.
@alex_prompter Alex Prompter on x
Zuckerberg hasn't posted on X in 3 years. Today he came back. Because when you finally have something to sell developers, you go where the developers are 🤷‍♂️ And the something: Meta is charging for AI now. Yup, $1.25 in, $4.25 out per million tokens, $20 in credits on every
@scaling01 @scaling01 on x
Muse Spark 1.1 scoring 59% on CyberGym this is worse than Opus 4.6 and Sonnet 4.6, but better than Opus 4.5 [image]
@uncledoomer @uncledoomer on x
hey mark quick question, how come youre posting about it on here? i thought your threads app was going to be the next twitter, bigger than twitter, even?
@scaling01 @scaling01 on x
Muse Spark 1.1 reaches 92.9% on CyBench, comparable with Claude Opus 4.6 (93%) [image]
@mikeisaac Rat King on x
you can tell mark really cares about something when he talks to #OldMedia for release day..... https://www.bloomberg.com/... [image]
@gergelyorosz Gergely Orosz on x
Remember Meta's track record in dev infra: TERRIBLE. In 2013, they bought a service loved by devs: Parse (backend-as-a-service for mobile apps). 2016: sudden shutdown, no migration path. 2023: released Llama, meant to be a series of open models. 2025: terminated this strategy
@alexandr_wang Alexandr Wang on x
Muse Spark 1.1 is SOTA on Harvey's Legal Bench, TaxEval, and MedScribe. It's cool to see that our model outperforms even Fable in a few areas :)
@alexandr_wang Alexandr Wang on x
Muse Spark 1.1 is ranked #4 on the Vals Index, ahead of GPT-5.5 and Grok 4.5
@gergelyorosz Gergely Orosz on x
This model looks impressive on paper. However, Meta has no history in dev tools, save for shutting down tools devs loved and used (Parse, 2013) It will take more than a single release to gain developer trust, and love. It takes years. I wonder if Meta will persist that long?
@kimmonismus @kimmonismus on x
Meta is releasing an agentic and coding model that can compete with GPT-5.5 and Opus 4.8? And on top of that, it is SOTA on HLE? Holy, that was not on my bingo card. Alexandr Wang cooked. What is happening today! [image]
@valsai @valsai on x
Meta just released Muse Spark 1.1 and is the new SOTA on MedScribe and TaxEval, taking the top spot from Fable 5 while being 10x cheaper and twice as fast. Meta currently holds the top 2 spots on TaxEval It is also the new #1 on Harvey's Legal Agent Bench, dethroning Grok 4.5 [im…
@rjccapital @rjccapital on x
I was promised excess compute
@alexandr_wang Alexandr Wang on x
if you wanna try computer use for yourself on your own machine, use this code! https://github.com/...
@agupta Ankit Gupta on x
Zuck being back on Twitter seems like an important sign of how seriously meta is taking this
@sriramk Sriram Krishnan on x
between Grok from @elonmusk / @mntruell and Muse Spark from @finkd this week very excited to see an ecosystem of frontier models in America.
@yasirmohiuddinn Yasir Mohiuddin on x
Muse spark 1.1 looks - interesting. Seems like meta woke up finally. [image]
@botaohao2 Botao Hao on x
Muse Spark 1.1 is out! It is a remarkable improvement over 1.0, and I'm proud that we built a strong RL system in such a short time to make it happen. It has been a lot of fun working with Alex and so many talented teammates on training Muse Spark 1.1!
@alignedai @alignedai on x
I am proud to have worked on post training Muse Spark 1.1. It is step wise improvement, giving us a REAL beachhead into the frontier! Especially happy to for the cost of intelligence plummeting and the inherent democratization of intelligence that stems from that. So proud of
@mrbye32 Yuri on x
We got: - GPT 5.6 Sol, Terra, Luna - Muse Spark 1.1 - Fable 5 (until 12 July) - Grok 4.5 - Gemini??? What's going on...
@hangsiin @hangsiin on x
Hmm... could you release Llama 5 as open source? Or maybe Muse Flash 1.1?...
@supbagholder @supbagholder on x
The speed $META released Muse Spark 1.0 and 1.1 back to back is showing real momentum. Muse Spark 1.0 -> 1.1 in ∼90 days. Things seem to be accelerating.
@jackda_techguy Jack Winterland on x
To use Muse Spark 1.1 on Meta AI, select thinking mode instead of the default instant mode. I'm pleased that my feed is full of Muse Spark post by @alexandr_wang, can't wait to try it out and see its capabilities!!! [image]
@robj3d3 Rob Hallam on x
Zuck's first tweet in 3 years being him dropping an Opus-level agentic coding model was not on my 2026 bingo card.
@omarsar0 Elvis on x
Muse Spark 1.1 available in new Meta Model API. Somewhere near Opus-4.8/GPT-5.5 level. 1M context window! Computer-use capabilities sound great: Write scripts when automation is faster, click when direct interaction is simpler, and generate batches of actions at each step.
@netcapgirl Sophie on x
ai race so crazy it has zuck posting on X
@biscuitweb3 @biscuitweb3 on x
meta muse spark 1.1 is more about cheap agents than model hype > $1.25 input > $4.25 output > 1m context > parallel subagents > strong tool use scores coding is still opus and gpt territory but agents that use tools, click through apps, run research and call APIs are the real
@tunguz Bojan Tunguz on x
Well, well, well - his first tweet since 2023, and second since 2012. Even he realized that unless you post here, no one is paying attention to what you are working on.
@mweinbach Max Weinbach on x
Zucc back on Twitter and Muse Spark 1.1 actually does look pretty good. Gonna need to try this in a bit
@avi_hacker1 @avi_hacker1 on x
Meta Muse Spark is interesting. Not the benchmark. It pushes agents toward normal software habits. Cheap APIs. Tool use. Subagents. Computer use. Hard part now is boring ownership when they actually act.
@andrewcurran_ Andrew Curran on x
Mark Zuckerberg returns to the platform to announce Muse Spark 1.1. [image]
@cybr4747 @cybr4747 on x
Grok 4.5 and Muse Spark 1.1 give me hope that SpaceX and Meta still have a chance to catch up. Yes, Anthropic and OpenAI are ahead but having Opus 4.8-class models, i.e. being just one generation behind, makes catching up possible. And with the intensity of both Elon and Zuck, if
@noctus91 @noctus91 on x
Good models are great. Good models with aggressive pricing are even better. Muse Spark 1.1 looks like a serious value play. [image]
@suchenzang Susan Zhang on x
x is now the new facebook, except you also get to enjoy the fruits of being cyberbullied by anon teenagers who are sharper, faster, and meaner than ever
@fanofaliens @fanofaliens on x
Meta should now build a standalone Meta AI desktop app for coding, agents, browser use, and computer control. All powered by Muse Spark. [image]
@thuris Shawn Thuris on x
I was impressed with Muse Spark in the few exchanges I had before this update. The new model will be worth a try.
@thomasscialom Thomas Scialom on x
Big day. Proud to be part of MSL as Muse Spark goes live, what an incredible team! Toolformer was only 3 years ago. Amazing to see how fast the field has moved, and to help bring more of this agentic direction into real products. Much more to come.
@nummanali Numman Ali on x
Muse Spark 1.1 looks to be a significant jump over 1 It rivals or beats other SOTA models in some benchmarks I'm curious given the short time period between releases All major labs are on the RSI curve so it's fair game to all! Competition is great for the consumer 👌
@kylebrussell Kyle Russell on x
Zuck is tweeting about the good model He's in the game
@abiggerspalash Palash on x
Zuck had to announce his new coding model on X because people on Threads thought Muse Spark must be some sort of a scented candle ✨🥰🕯️
@itsmeajaykv @itsmeajaykv on x
Yesterday we got Grok 4.5, today we are getting Muse Spark 1.1 Comparable with Opus 4.8 (max) and GPT 5.5 (xhigh) on frontier territory. Comes with 1M context ! Looks like Gemini 3.1 pro is falling behind, has a lot of catching up to do.
@natliml Nathaniel Li on x
We're releasing the Muse Spark 1.1 Evaluation Report! We've substantially improved the robustness and cyber safeguards for Muse, as we cannot rule out it meeting our “high” capabilities threshold in cybersecurity. [image]
@mmatt14 @mmatt14 on x
So Zuck is going for pay to use on Muse Spark Imho it's not a problem
@zephyr_z9 @zephyr_z9 on x
Much more competitive on pricing This model will see great adoption [image]
@thestalwart Joe Weisenthal on x
Other than a joke post about the launch of Threads, first time he's said something on here in over a decade
@arena @arena on x
Muse Spark 1.1 by @AIatMeta is now in the Agent Arena! In Agent Arena, we measure models on millions of real-world, long-horizon agentic tasks from a global community of users. Models can access web search, filesystem, and terminal tools to complete complex workflows. The [image]
@iterintellectus Vittorio on x
1) what? [image]
@jerrycap @jerrycap on x
Zuck is tweeting?!
@jack_w_rae Jack Rae on x
Another release this week🔥 Muse Spark 1.1 is a considerable upgrade in capability, especially for agentic tasks. Plus... It's available via API! We see this model as a workhorse: fast and very reasonable priced, with strong all-round capability.
@fanofaliens @fanofaliens on x
Mark Zuckerberg returned to 𝕏 after years to announce Meta's new AI model, Muse Spark 1.1. It's built for coding, tool use, computer control, long tasks, and 1M context. We are so back. [image]
@synthwavedd Leo on x
For my second “lmao wtf” moment of this week... Meta just announced Muse Spark 1.1, and it's also a frontier level model competing with Opus 4.8 and GPT-5.5 Zuck AND Elon are back [image]
@edwardsun0909 Zhiqing Sun on x
Excited to share Muse Spark 1.1 with you! It's our first model available in the API, built to excel in agentic and coding. I'm incredibly proud of what the team has achieved in such a short time, from 1.0 to 1.1. The momentum is high, and we're still training larger models! [imag…
@yuchenj_uw Yuchen Jin on x
Didn't know Zuck had a Twitter account. His last tweet was in 2023. Pretty much confirms X is the best place to talk about AI.
@aiatmeta @aiatmeta on x
We're excited to introduce Muse Spark 1.1, a significant upgrade from the first Muse Spark model we released earlier this year. Along with this release, we are launching a public preview of the new Meta Model API where developers can access Muse Spark 1.1. The model is also [imag…
@shengjia_zhao Shengjia Zhao on x
Today we are launching Muse Spark 1.1, an upgrade to muse spark 1 that greatly improves agentic, coding, multimodal, and computer use capabilities. We're also launching the Meta Model API in public preview. https://ai.meta.com/... [image]
@rihardjarc Rihard Jarc on x
[...] So $META is starting to ship on the AI model front, while at the same time they seem to have locked in the supply chain to double their AI capacity to 14GW in 2027. Yet the market thinks this is negative. $META also seems willing to sell excess capacity to the market, whe…
@danielamitay Daniel Amitay on x
I've been using Muse Spark for a while now and it's truly a great model.
@alexandr_wang Alexandr Wang on x
3/ computer use: it can operate desktop, browser, and mobile. muse spark 1.1 is trained to know when to automate and when to use interfaces directly. it scripts when that's faster, clicks when direct interaction is simpler, and batches actions at each step. [video]
@alexandr_wang Alexandr Wang on x
2/ agentic performance and tool use are where we're leading. it has strong performance on long-running agentic tasks with a 1M token context window with active context management, and can delegate execution across parallel-running subagents [image]
@aiatmeta @aiatmeta on x
Straight from @finkd — Muse Spark 1.1 is live.
@alexandr_wang Alexandr Wang on x
muse spark 1.1 is out! our strongest model yet for agentic and coding work, now available in the meta model api. more to come. proud of the team on this one!
@finkd Mark Zuckerberg on x
(2) Muse Spark 1.1 is strongest at agentic performance, tool use, and computer use. It does well on long-running tasks with 1M token context window, can delegate execution to sub-agents running in parallel, and is trained to use computer interfaces on desktop, mobile, or browser.…
Hyung Won Chung Hyung Won Chung on linkedin
We're releasing Muse Spark 1.1 today with a big step up in agentic capabilities https://lnkd.in/... Also releasing the API, and it is really, really cheap! …
Wen Shi Wen Shi on linkedin
Big milestone at Meta today. We're launching Muse Spark 1.1 — a meaningful step up in coding, agentic reasoning, and multimodal understanding …

Chronicles