Muse Spark 1.3

DataCamp's hands-on tutorial confirms that Meta released Muse Spark 1.3 on September 2, 2026, and that developers can adopt it with a one-line model-ID change in Muse Code, using the same endpoints, SDK, and pricing as Muse Spark 1.2. The article documents that the model is multimodal (text, images, video, files) with The same piece adds practical developer findings from three real coding tasks run on an intentionally broken open-source project: across six runs, Muse Spark 1.3 was cheaper on two tasks and 38% more expensive on the third, netting a 12% cost increase; model completions fell 23% and 32% on the tasks where 1.3 won, but

Shattered.io's September 5, 2026 article focuses on the headline efficiency claim around Meta Muse Spark 1.3: roughly 20% fewer tool calls and about 25% fewer tokens than Muse Spark 1.2 to complete equivalent agentic coding tasks. The piece confirms the September 2, 2026 release and availability in Muse Code and throug The article explicitly flags that the 20%/25% efficiency figures come from internal comparisons run by Meta engineers rather than independent third-party benchmarks, recommending teams treat them as a starting point for their own testing. A standard access tier is mentioned at $1.25 per (unit truncated in excerpt). Met

Flowtivity's analysis reports that Meta released Muse Spark 1.3 on September 2, 2026 from Meta Superintelligence Labs, claiming it as Meta's biggest jump yet on coding and agentic tasks: 75.4% on DeepSWE 1.1 for end-to-end agentic software engineering, 88.8% on Terminal-Bench 2.1, 59.4% on SWEAtlas CodeBase QnA, and 98 The piece frames Muse Spark 1.3 as Meta's first closed, directly monetised frontier line shipping on a four-week cadence, with same-day availability in Muse Code and the Meta Model API and paid developer access opening September 3. It positions the release as pressuring rivals on price while noting that, according to M

BenchLM profiles Muse Spark 1.3 as a proprietary reasoning model released on September 2, 2026, with an API model ID of "muse-spark-1.3", a 1M-token context window, multimodal inputs (text, image, video, document), and text output, attributing these specs to a "Meta Muse Spark 1.3 model page." The page is explicitly la For pricing, BenchLM lists $1.25 per million input tokens and $4.25 per million output tokens, with $0.15 cached input and a $2.75 blended figure, all sourced from upstream tracking pages rather than a Meta rate card — consistent with the third-party-sourced pricing flagged in the aireleasetracker excerpt. Fields such

Meta's Muse Spark 1.3 (model ID meta/muse-spark-1.3) was released on OpenRouter on September 2, 2026, as a multimodal reasoning model targeting long-running agentic, multi-agent, and coding workflows. It supports text, image, video, and document inputs with text output, though audio understanding is flagged as not full OpenRouter routes all requests directly to Meta with no routing decisions. Measured performance shows 91 tokens per second P50 throughput, 2.65s P50 latency, and 100.00% uptime. The model scores 61.4% on the AutoExacto Bench (vgi bench 02823add), with an average tool call error rate of 1.42% and a structured output err

Meta released Muse Spark 1.3 on Wednesday, September 2, 2026, as an update that the company says significantly improves performance in coding and agentic tasks. In an interview with Axios, Meta AI chief Alexandr Wang described the model as "very competitive with frontier models" and tied the improvements to upcoming Me For developers, the practical signal in this Axios article is the explicit Meta-attributed framing of Muse Spark 1.3 as a coding- and agentic-focused step, with Wang directly linking the release to Mark Zuckerberg's stated personal-agent ambitions. No API identifiers, pricing, context window, or benchmark figures appea

AI Release Tracker's model page confirms Meta released Muse Spark 1.3 on Wednesday, September 2, 2026, 23 days after Muse Glimmer, with a 1M token context window. It lists benchmark coverage across DeepSWE 1.1, SWEAtlas CodeBase QnA, Terminal-Bench 2.1, JobBench, DeepSearchQA, Agentic IF Index, and MRCR (plus five more The tracker lists third-party rates of $1.25 input / $4.25 output per 1M tokens, explicitly sourced from openrouter.ai on September 21, 2026 rather than Meta's own list price. Pricing is presented as a neutral cross-reference with a clear fetch timestamp, making the page a useful aggregator for specs and benchmark rank

The LLM Stats aggregator page for Muse Spark 1.3 (dated September 2, 2026) places the model in capability tiers including Top 2% Long Context (1 of 119), Tool Calling (2 of 192), Reasoning (7 of 361), and Coding (9 of 265), with an LLM Stats Score of about 55.1 against a blended price around $0.10 per million tokens. B Quality tracker data shows 25 votes within the tracking window, and the page is gated behind a human-verification step. The blended price figure is aggregator-computed rather than a Meta-published API rate, and several benchmark rows are truncated in the available excerpt. The page does not mention OpenRouter despite t