Muse Spark 1.3 Contributor

Meta shipped Muse Spark 1.3 on September 2, 2026 as an update to its API and to Muse Code, Meta's agent harness, accompanied by documentation and model-catalog changes that landed without a dedicated keynote moment. Meta AI chief Alexandr Wang told Axios the release is part of the groundwork for personal agents that ca Beneath that framing, Muse Spark 1.3 ships in two model IDs: muse-spark-1.3 as the standard version, and muse-spark-1.3-contributor as a heavily discounted variant that trains on user prompts and completions. Which model a piece of software actually calls is determined entirely by a short string parameter in an API req

Meta shipped Muse Spark 1.3 on September 2, 2026, with the upgrade path fitting on one line: change the model ID, since the same endpoints, SDK, and pricing apply. According to Meta, the new model finishes the same task with roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2, numbers that come from c Across six runs on three coding tasks, 1.3 was cheaper on two and 38% more expensive on the third for a net 12% cost increase over the set, with model completions falling 23% and 32% on the two tasks where 1.3 won, lining up with Meta's tool-call claim. Uncached input fell on all three tasks but never by 25%, and Muse

The Shattered.io coverage of Meta's Muse Spark 1.3 launch (September 2, 2026) documents the model's positioning as an agentic and coding model built for long-horizon tasks involving multi-step planning, tool calling, and self-checking, and confirms availability in both Muse Code and the Meta Model API. The piece restat The article confirms two access tiers for Muse Spark 1.3 — standard access at $1.25 per million tokens and a contributor tier at lower pricing where Meta can use the traffic — and frames the practical impact for teams running coding agents at scale, where reduced tool-call latency and token spend translate directly int

The Medium analysis by Gnana Prasath confirms Meta's September 2, 2026 Muse Spark 1.3 release and frames it as a long-horizon agentic and coding model with text, image, and video input, text output, and a roughly one-million-token context window, with the previously available reasoning modes (including xhigh) shipping The Contributor tier is documented at $0.10 input and $0.20 output per million tokens, with multimodal input and a 1M-token context window, and Meta's highest reasoning mode (max) is flagged as not generally available at launch, pending additional safety work, so launch-day claims tied to max should be separated from w

Meta shipped Muse Spark 1.3 on September 2, 2026, making it available immediately inside Muse Code, Meta's coding agent, and through the Meta Model API as a drop-in upgrade where endpoints, SDKs, and prices stay the same. The release targets long-running, multi-agent workflows where a model has to track context across Muse Spark 1.3 is described as a proprietary, closed-weights multimodal reasoning model that accepts text, images, and video across a context window of about one million tokens, with previously available reasoning settings including xhigh carried over and the max reasoning setting in limited preview for partners. Muse

Meta's press blog described Muse Spark 1.3 as a model optimized for competitive coding performance, with Meta's own benchmark table showing it scoring 75.4 on DeepSWE v1.1 ahead of Claude Opus 5's 74.0 and GPT-5.6 Sol's 73.0, and tying GPT-5.6 Sol at 88.8 on Terminal-Bench 2.1 while passing Opus 5's 86.7. On two of nin The launch includes a contributor tier that is up to 21x cheaper in exchange for letting Meta train on usage, and Alexandr Wang later confirmed on X that the entire Spark 1 through 1.3 line was built with one target in mind: Muse, Meta's personal-agent product that shares the name for a reason. Meta Superintelligence L

Meta released Muse Spark 1.3 on September 2, 2026, as the latest model in the Muse Spark family, designed for long-running AI agents, coding, tool use, long context, and complex workflows rather than single-turn chat. According to Meta, the new version is more efficient than Muse Spark 1.2, using roughly 20% fewer tool Beyond chat, Meta's examples show Muse Spark 1.3 applied to practical deliverables such as turning an Excel file of constituent feedback into a structured PDF with talking points, key numbers, and district-specific material for a board meeting. The release targets software engineering, research and web browsing, comput

The OpenRouter catalog entry for Meta Muse Spark 1.3 Contributor describes it as the cost-efficient contributor tier of Meta's multimodal reasoning model, intended for experimentation, learning, and early-stage agentic, multi-agent, and coding workflows. It tracks information across extended tasks, works through confli Performance metrics on the listing include a P50 latency of 4.93 seconds and throughput of 59 tokens per second, with a Meta provider tool-call error rate averaging 2.96%. Cache read pricing is listed at $0.002 per million tokens, and the weighted average input price actually paid by customers is $0.01908 per million t