Meta Muse Spark 1.1 Beats Opus, Grok 4.5 and Gemini On New Theoretical Computer Science Benchmark
Meta Muse Spark 1.1 topped a new theoretical computer science benchmark, outperforming rival models including Anthropic Opus, X.AI Grok 4.5 and Google Gemini on a test of graph induction reasoning. Researcher Dimitris Batzoglou posted the results Friday, noting the evaluation comprises 64 problems where models must generate logical formulas to describe designated nodes across multiple graphs.
The performance extends Muse Spark 1.1’s recent rankings on scientific coding and agent evaluation suites. Batzoglou stated the benchmark, described as a spotlight paper at the ICML 2026 conference, is now publicly available. The results follow days of independent testing that placed the model near or ahead of earlier frontier releases.
From the sources (25 posts)
@charlesrollet1RT @alexeheath: Meta AI chief @alexandr_wang told me the company sees its new API for selling AI models as a real business, not just a lear…
@garrytanRT @alexeheath: Meta AI chief @alexandr_wang told me the company sees its new API for selling AI models as a real business, not just a lear…
@alexeheathMeta AI chief @alexandr_wang told me the company sees its new API for selling AI models as a real business, not just a learning exercise. Tokens are “gigantic and growing very, very quickly,” he told me, and capturing even a slice of that
@kimmonismusI do not think anyone saw this massive comeback from Meta coming. And now they are even receiving high praise from SemiAnalysis. Just to be clear: I have the utmost respect for SemiAnalysis and Dylan Patel. They do truly outstanding resear
@alexandr_wangRT @robertcourson: I am very impressed with Muse Spark 1.1 for UI/UX. Especially for the price. This was the result of 1 prompt to make a…
@alexandr_wangRT @mweinbach: Let's see how Muse Spark 1.1 is!
@alexandr_wangRT @emdashsh: Muse Spark 1.1 is now available in Emdash with OpenCode.
@alexandr_wangRT @eashish93: I was testing Muse Spark 1.1 and honestly, the result is pretty impressive. Built from planning to completion in a single p…
@mweinbachJust because this got retweeted, Muse Spark 1.1 impressions are it's actually pretty good This fills the gap for me that Kimi K2.6 or GLM 5.2 would, good enough for most things, pretty quick, very cheap (more efficient than those models)
@alexandr_wangRT @mweinbach: Just because this got retweeted, Muse Spark 1.1 impressions are it's actually pretty good This fills the gap for me that Ki…
@wesrothVals AI’s latest results place Meta Muse Spark 1.1 among the strongest new agent models. It is now ranked #1 on MedScribe and TaxEval, taking the top spot from Fable 5, while reportedly being 10x cheaper and twice as fast. Muse Spark 1.1
@therundownaiRT @rowancheung: Many people were doubting Meta's position in the AI race. Yesterday, they dropped Muse Spark 1.1, now one of the stronges…
@rohanpaul_aiRT @rohanpaul_ai: Meta is so back in the AI coding race with Muse Spark 1.1, using cut-rate pricing to pressure OpenAI and Anthropic in age…
@alexandr_wangmuse spark is able to do end-to-end tasks based on short video instructions
@alexandr_wangRT @aicodeking: Meta Muse Spark 1.1 is a great model. Awesome at frontend and everything. It still lacks at some places like overwriting fi…
@alexandr_wangRT @CuriousRefuge: So @Meta has officially entered the AI image and video space, and we're excited to share our early test results. Muse Vi…
@teortaxestexMuse Spark 1.1 is surprisingly close to Grok 4.5 on many high-signal evals This is the current top on CritPT
@yunta_tsaiIt is interesting that Muse Spark 1.1 has 1M context length but did not perform as well as Grok 4.5 at 500K context length. Sometimes I feel people chasing context length without scrutinizing every token inside the context are wasting memo
@alexandr_wangmuse spark 1.1 can be even better than opus 4.8 at 20% of the cost
@denny_zhoumuse spark 1.1 passed elon bench
@ofirpressRT @alexandr_wang: muse spark 1.1 is ahead of gpt-5.6 on SciCode
@alexandr_wang@teortaxesTex this isn’t true we’re going to increase access to muse spark incl openrouter
@alexandr_wangmuse spark 1.1 outperforms opus, grok 4.5, and gemini on a new challenging finite model theory / theoretical cs eval
@jack_w_raeRT @alexandr_wang: muse spark 1.1 outperforms opus, grok 4.5, and gemini on a new challenging finite model theory / theoretical cs eval
@s_batzoglouI benchmarked the new models (Sol, Terra, Luna, Fable 5, Meta Muse Spark 1.1, Grok 4.5) on an induction reasoning task. This is an updated table from yesterday, and my benchmark is described as spotlight in ICML '26. How to read the tabl