Muse Code vs. the field.
Price and capability against Claude Opus 5, OpenAI Codex, DeepSeek V4, GLM-5.2, and MiniMax M3 — from vendor price sheets and published benchmarks, as of August 7, 2026.
API pricing, per 1M tokens
Cached input is what agentic coding mostly pays for — repository context re-sent every turn.
Coding-agent capability
Agent benchmarks measure the model inside its harness. Scores from Meta’s published evaluation and vendor-reported results; missing cells mean no published score.
Terminal-Bench 2.1, the full field
Meta’s published chart adds context the table can’t: Muse Spark 1.2 in Muse Code (82.9%) sits second behind Claude Opus 5 max in Claude Code (86.7%), ahead of GPT 5.6 Terra max in Codex (81.8%), Grok 4.5 high in Grok Build (81.6%), and Gemini 3.6 Flash high in Antigravity CLI (78.9%) — and a full 6.7 points above its own predecessor, Muse Spark 1.1 in mini-swe-agent (76.2%).
On price, two different fights
Muse standard ($1.25/$4.25) undercuts Claude Opus 5 ($5/$25) by 4–6x while sitting above the Chinese open-weight trio. The Contributor tier ($0.10/$0.20) drops below even DeepSeek V4 Flash ($0.14) on input — the price of sharing your data.
On capability, second place — closely
On Terminal-Bench 2.1, Muse Spark 1.2 + Muse Code (82.9%) trails Claude Opus 5 (86.7%) and edges Codex running GPT-5.6 Terra (81.8%), with Grok Build (81.6%) and Antigravity CLI (78.9%) behind. On DeepSWE 1.1 the gap is wider (59.3% vs. 65.0%). Frontier-adjacent, at a fraction of frontier price.
The cache column decides real bills
Agent sessions re-send repository context every turn. Muse Contributor’s $0.002 cached input is the lowest figure on the table; DeepSeek V4 Flash’s $0.0028 is the only close rival. For long sessions, this row matters more than headline input price.
Sources: Meta Model API pricing and evaluation report (research.meta.ai); OpenAI (GPT-5.6 rates after the July 30, 2026 cut), Anthropic, xAI (Grok 4.5, July 8, 2026), Google (Gemini 3.6 Flash, July 21, 2026), DeepSeek (V4 Flash public beta, July 31, 2026), Z.AI, and MiniMax published price sheets; third-party pricing trackers, checked August 7, 2026. Benchmark conditions differ across vendors — treat scores as directional. Prices change; verify on each vendor’s page before budgeting.