DataCamp's hands-on developer tutorial for Muse Spark 1.3 found that Meta's efficiency claims held up only partially under independent testing. Across 6 runs on 3 coding tasks against Muse Spark 1.2, Muse Spark 1.3 was cheaper on 2 tasks but 38% more expensive on the third, netting a 12% cost increase. Model completion The tutorial surfaces several operational details for developers: Muse Code defaults to muse-spark-1.3-contributor, which trains on user code unless manually switched (training is opt-out, not opt-in). The ultra reasoning level exists in the CLI and in-session picker but is blocked by a named feature gate on the backen

Meta released Muse Spark 1.3 on September 2, 2026, from Meta Superintelligence Labs, positioning it as the company's most significant update of the year for AI agent development. According to figures cited from Meta AI Research, the model achieves 75.4% on DeepSWE 1.1 for end-to-end agentic software engineering, 88.8% Meta's internal comparisons show Muse Spark 1.3 finishing coding work with approximately 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2, though these efficiency gains were not independently verified in the source. Strategically, Muse Spark 1.3 marks Meta's first closed, directly monetised frontier model

Muse Spark 1.3 was released by Meta on Wednesday, September 2, 2026, 23 days after Muse Glimmer, featuring a 1 million token context window. Benchmark results tracked include DeepSWE 1.1 (75.4%, best published score among tracked models), SWEAtlas CodeBase QnA (59.4%), and Terminal-Bench 2.1 (88.8%, ranking 3 of 30, be Provider pricing is listed at $1.25 input and $4.25 output per million tokens (USD), fetched from openrouter.ai on September 21, 2026, representing third-party aggregator rates rather than Meta's official list price. The model is served by Meta. DeepSWE 1.1 is highlighted as achieving the best published score of all tr

Muse Spark 1.3 ranks on the LLM Stats Score composite at 54.3, with a blended price of $1.39 per million tokens, placing it between DeepSeek-V4.1-Flash ($0.24, score 51.8) and GPT-6 Astra ($11.9, score 59.6) on the cost-efficiency axis. On specific benchmarks, the model scores 0.98/1 on MRCR v2 (8-needle) in both the 2 The LLM Stats Quality Tracker shows Muse Spark 1.3 at -0.07σ over a 7-day window, trending as "Improving" with 59 votes. Across tracked datasets, the model demonstrates strong long-context retrieval consistency (0.98 at both 256K–512K and 512K–1M token ranges) and competitive agentic performance on GDPval-AA. The compo