Zuckerberg unveiled Muse Code on Wednesday, Meta's first AI coding agent, but the company's own test results tell an uncomfortable story. Anthropic's Claude Opus 5 beat it across all four benchmarks Meta published. The gap matters less than the fact Meta released the charts anyway, betting on price and convenience over raw performance.

Muse Code runs on Muse Spark 1.2, a coding-focused model update designed to handle complete software engineering tasks. On Terminal-Bench 2.1, which tests 89 real-world jobs spanning system repair and security, Muse Spark scored 82.9%. Claude Opus 5 hit 86.7%. OpenAI's Codex came in third at 81.8%.

The next test got worse. DeepSWE 1.1 contains 113 coding tasks with internet access disabled during grading. Muse Spark 1.2 dropped to 59.3%, landing behind both competitors. Meta then ran its own benchmark using 440 real pull requests from internal engineers. The model scored 70.6%, roughly nine points below Opus 5.

The Model Meta Didn't Mention

Here's where the story gets interesting. OpenAI sells a stronger model called Sol that tops the independent Terminal-Bench 2.1 leaderboard at 89.5%. Claude Opus 5 follows at 89.1%. Both figures beat the 86.7% number Meta reported for Opus 5. The company picked a weaker setting of its strongest rival and still finished second.

Against Sol, the actual leader, Muse Spark 1.2 trails by 6.6 points rather than the 3.8-point gap Meta's charts suggest. Meta included Sol in just one place, a GPU kernel task where it improved the baseline by 71.2% compared to Muse Spark's 68.7%.

The strategy is transparent enough. Meta is selling a cheaper tool rather than a better one. The company's own engineers demonstrated the model can run complex optimization tasks for 24 hours on NVIDIA Hopper hardware, making thousands of tool calls and finding real speedups. That capability matters for teams on a budget. But the benchmark results, even when presented by Meta itself, show Anthropic still owns the coding agent space.

This article is informational only and does not constitute financial or investment advice.