Wang Says Muse Spark 1.3 Closes Meta's Agent Gap With Frontier Labs
Meta AI chief Alexandr Wang told Axios and Bloomberg that Muse Spark 1.3 is competitive with Claude Fable 5.1 on agentic work, priced unchanged from 1.2, and aimed at personal agents that run continuously.
Classy aggregates publicly available material; we did not conduct a private interview.
What changed
On 2 September 2026 Meta released Muse Spark 1.3 through Muse Code and the Meta Model API. Chief AI officer Alexandr Wang told Axios the update is "very competitive with frontier models" and said in a Bloomberg interview that it represents Meta's largest performance jump so far in the Muse Spark line. The model ships with a one million token context window, multimodal input, and reasoning tiers controlled through a reasoning_effort parameter, with a maximum reasoning mode still undergoing safety testing.
Wang tied the release directly to Meta's personal agent roadmap. He told Axios that usability improvements "will be really helpful for things that Mark [Zuckerberg] has talked about on earnings calls, like personal agents that can work 24/7 on your behalf." Meta kept API pricing at the same level as Muse Spark 1.2, which Wang described as aggressive.
Why it matters
Meta is trying to prove it can ship frontier tier models on a monthly cadence without raising API prices. For product teams building always on assistants, Wang's framing matters more than a single benchmark row: the company is optimizing for long horizon tool use, coding, and retrieval inside one million token threads rather than one shot chat.
If Muse Spark 1.3 holds up in independent evals, it gives Meta a credible closed weight option for agent stacks that today default to Anthropic or OpenAI. Wang claimed the model beats GPT 5.6 Sol on coding and matches Claude Fable 5.1 on agentic tasks in Meta's internal comparisons, but those figures need third party confirmation before they drive procurement.
Who is affected
Applied AI leads evaluating a third frontier vendor for agent products. Social and messaging teams inside Meta that will route consumer agents through the same model family. Developers already running trillions of tokens per week on Muse Spark APIs, per Wang's Bloomberg comments. Competitors watching whether Meta's Scale AI backed lab can sustain monthly releases without safety regressions.
What to do next
Run your agent eval harness on Muse Spark 1.3 with the same tool schemas and cache patterns you use for Fable 5.1 and GPT 6 Astra. Compare cost per completed task, not list price alone, because Wang emphasized token efficiency gains of roughly 25 percent versus Muse Spark 1.2.
What to watch
Meta's rollout of the maximum reasoning mode after additional safety testing. Independent replication of DeepSWE and long context MRCR scores Meta published for 1.3. Any consumer agent features Meta ships on Facebook, Instagram, or Meta AI that depend on 1.3 level tool reliability.
Sources
- Primary. Axios, Meta debuts Muse Spark 1.3 as personal agent work continues (2 September 2026). Wang quotes on frontier competitiveness, personal agents, and unchanged pricing.
- Secondary. SiliconANGLE, Meta says it has caught up with Anthropic and OpenAI with Muse Spark 1.3 (2 September 2026). Bloomberg interview claims on coding versus GPT 5.6 Sol and parity with Fable 5.1.