Muse Spark 1.2 Contributor

On August 5, 2026 (Eastern Time), Mark Zuckerberg announced Meta's Muse Spark 1.2 model via an X post, launching it alongside the beta release of Muse Code, a terminal-based coding agent designed to handle full software engineering tasks across large repositories including planning, code writing, and result validation. The report details the Contributor tier pricing for Muse Spark 1.2: input at $0.10 per million tokens and output at $0.20 per million tokens — undercutting DeepSeek Flash's pricing on the same day DeepSeek signaled an upcoming substantial API price increase. The Contributor plan's condition is that users agree to let M

Meta Superintelligence Labs released Muse Code in beta alongside Muse Spark 1.2 on August 5, 2026, with the article reporting that the model reaches 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE v1.1, up from 76.2% and 53.0% for Muse Spark 1.1 according to Meta's own evaluations. The Contributor tier is offered at $ Muse Code is a terminal-only agent for macOS and Linux (no Windows build at launch) that installs via a single curl command and ships with three default skills — /plan for approval-gated planning, /grill for stress-testing the plan, and /goal for working toward a stated objective. The runtime uses a local append-only e

Meta released Muse Code (beta) and Muse Spark 1.2 on August 5, 2026, with Muse Code available as a terminal coding agent for macOS and Linux powered by Muse Spark 1.2. Developers Digest reports that the model and the harness were co-trained — Muse Spark 1.2 was trained inside its own agent runtime so the model's behavi The contributor tier lands in the same pricing band as DeepSeek V4 Flash ($0.14/$0.28) and below GPT-5.6 Luna's post-cut ($0.20/$1.20) rates, and makes the data-for-discount trade explicit in a way most vendors keep implicit. The article points to official sources including the Meta AI Research announcement, the Muse C

Ginger Labs provides a technical analysis of the Muse Spark 1.2 Contributor API, framing it as an evaluation lane with distinct operational characteristics. The article notes that Contributor reportedly discounts the API in exchange for permission to use submitted prompts and completions to train future models, and rec On the technical side, the analysis references Muse Spark 1.1's documented capabilities as context for the family — including a 1-million-token context window, agentic delegation, context compaction, tool use via an OpenAI-compatible interface, and multimodal reasoning — and characterizes Muse Spark 1.2 as the coding-f