A Japanese technical dissection published on note.com on August 6, 2026 documents Meta's August 5 announcement of Muse Code and the underlying Muse Spark 1.2, drawing only on Meta's official blog and evaluation methodology report. It describes Muse Code as a resident background coding agent and covers the demo of gener The piece is auto-translated from Japanese, which the site warns may miss nuances and authorial intent, and benchmark figures are restated second-hand from Meta rather than independently verified. It also does not name AIHubMix as a provider, limiting its relevance to gateway or routing questions. The architectural and

Vorp Labs' August 10, 2026 release review treats Muse Spark 1.2 as Meta Superintelligence Labs' coding-focused update to Muse Spark 1.1, shipped alongside Muse Code, a beta terminal coding agent co-trained with the model. Standard API pricing remains $1.25 input, $0.15 cached input, and $4.25 output per 1M tokens with The most notable commercial detail is the new muse-spark-1.2-contributor tier priced at $0.10 input and $0.20 output, a 12.5x input and 21.25x output discount relative to standard rates in exchange for permission to train future Meta models on user prompts and completions, capped at 60 requests per minute against the s

Artificial Analysis published a deep benchmark analysis of Muse Spark 1.2 on August 5, 2026, with the lab receiving pre-release access from Meta. Spark 1.2 (xhigh) lands at 54 on the Artificial Analysis Intelligence Index, up 3 points from Spark 1.1 (51) and 11 points from Spark 1.0 (43), placing it effectively tied wi At unchanged $1.25 input and $4.25 output per 1M token pricing, Spark 1.2 costs $0.40 per Intelligence Index task, more expensive per task than Grok 4.5 ($0.37) and GPT-5.6 Sol ($0.39) but cheaper than GPT-5.6 Terra, Kimi K3, and GPT-5.5; cost rose from $0.29 on Spark 1.1 due to 53% higher input and 36% higher output t

The AI Release Tracker entry for Meta Muse Spark 1.2 documents its August 5, 2026 release, arriving 27 days after Muse Spark 1.1, and confirms a 1M token context window. Reported benchmark scores include BullshitBench v2 at 50% (27 of 77), DeepSWE 1.1 at 59.3% (15 of 27), and Terminal-Bench 2.1 at 82.9% (14 of 30), wit The tracker frames Spark 1.2 as part of an unusually dense frontier release calendar, with the model becoming the most recent entry on the site's August 2026 ledger at the time of publication. Coverage includes standard comparison links against Muse Spark 1.3, GPT-6 Astra, and Claude Fable 5.1. Pricing is explicitly fl

BenchLM's catalog page for Muse Spark 1.2 reports a composite capability score of 70.4/100, with Knowledge ranking 10th (95th percentile), Agentic 17th of 153, and Coding 21st of 152, against a field median of 56.2. Published pricing is $1.25 input and $4.25 output per 1M tokens with $0.15 cached input and a $2.75 blen The page explicitly flags Muse Spark 1.2 as superseded by Muse Spark 1.3, signaling that the tracked benchmarks are no longer current within Meta's own lineup. It also notes the model is "particularly effective for knowledge-intensive tasks like research, analysis, and factual Q&A," while coding ranks as its lowest eli