Meta’s Claude Code clone is INSANELY cheap

Summary
Theo Browne (t3.gg) reviews Meta’s release of the “Muse Code” terminal coding agent and its underlying “Muse Spark 1.2” coding model. He tests its performance, speed, integration, and pricing across several coding benchmarks and tasks in his open-source t3code repository.

What is shown

• [01:03] Mark Zuckerberg’s announcement post on X regarding the launch of Muse Code in beta powered by Muse Spark 1.2.
• [04:33] Review of benchmarks comparing Muse Spark 1.2 against models like Claude Opus 5, GPT-5.6 Terra, and Grok 4.5 across Terminal-Bench 2.1, DeepSWE 1.1, and Meta’s Internal Coding Bench.
• [07:22] Examination of Meta Model API pricing: standard pricing ($1.25/M input, $0.15/M cached input, $4.25/M output) versus the discounted “contributor tier” ($0.10/M input, $0.002/M cached input, $0.20/M output).
• [08:59] Artificial Analysis dashboard metrics comparing token costs and OpenRouter throughput speeds (showing ~191 tokens/sec for Muse Spark 1.2).
• [13:30] Live demonstration of muse CLI running inside the t3code codebase, importing Claude Code configurations and skills.
• [18:16] Generation of a 3D and 2D “fishslop” browser game directly from prompt instructions using Muse Code.
• [25:28] Muse Code encounters severe rate limits and hallucinates heavily, misidentifying “Muse” as Google’s “Antigravity” CLI.
• [37:50] Running a PR audit across 222 pull requests on GitHub using parallel sub-agents, outputting an interactive HTML audit dashboard.

Claims & numbers

• Mark Zuckerberg claims Muse Code runs specialized background agents that stay active across sessions and fan out to sub-agents working in isolated worktrees (01:16, 05:44).
• The presenter notes standard API pricing for Muse Spark 1.2 is $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens, while the contributor tier drops to $0.10 input, $0.002 cached input, and $0.20 output (07:22, 07:35).
• On OpenRouter, the presenter notes Muse Spark 1.2 achieves an average throughput of ~191 tokens per second (with peak measurements shown up to 316 tokens/sec) compared to ~30–50 tokens/sec for GPT-5.6 Sol (12:15, 12:42).
• The presenter claims Artificial Analysis scores Muse Spark 1.2 at 54 on the Intelligence Index, tying with GPT-5.5 and Grok 4.5 (08:00, 08:28).
• The presenter shows his full audit of 222 pull requests finished in under 5 minutes and cost roughly 10 cents on the contributor tier (42:04, 42:13).

Notable quotes

• [01:10] “This is their clone of Claude Code, and it’s powered by Muse Spark 1.2, which is their new coding-focused model.”
• [07:43] “That is a 10 to 20x price gap. They’re effectively giving out the contributor version for free, because they need training data so goddamn badly.”
• [34:42] “Well, the obvious reason, like the number one thing that would make someone want to use this model, is that they work at Meta.”

Assessment
This is a hands-on review and live testing video by an independent developer evaluating a newly released product. The demos and benchmarks are run live on-screen, capturing authentic shortcomings such as severe API rate limiting, 3D rendering control glitches, and context hallucinations alongside its high generation speeds.

Described by gemini-3.8-flash on 2026-10-13 from the video’s audio and frames.