An Open-Source Model Just Beat GPT-5.4 and Opus 4.6 on the Hardest Coding Benchmark

GLM-5.1 from Zhipu AI scored 58.4% on SWE-bench Pro. GPT-5.4 scored 57.7%. Claude Opus 4.6 scored 57.3%. That's the first time an open-weight model has ever topped a major coding benchmark against the best proprietary models.

The specs matter. GLM-5.1 is a 754B-parameter mixture-of-experts model with 40B active parameters, 200K context window, MIT license. That last part is the headline. MIT license means anyone can deploy it, fine-tune it, build products on it. No usage restrictions. No revenue thresholds. No terms that change at a founder's whim.

I want to be careful about overstating a 0.7% margin on a single benchmark. SWE-bench Pro is the hardest variant of the most respected coding evaluation, so it's not a cherry-picked leaderboard. But benchmarks capture a slice of capability, not the whole picture. I've used both Opus 4.6 and GPT-5.4 extensively in production, and a fractional benchmark difference rarely maps to a noticeable quality gap in daily use.

What's more interesting is the trend. Zhipu AI became the first publicly traded foundation model company after their HKD 4.35B Hong Kong IPO in January 2026. Chinese labs, Zhipu, Alibaba, Moonshot AI, DeepSeek, now hold most of the top open-weight positions. Google's Gemma 4 31B broke into the top 5 as well. The competitive pressure on the proprietary model providers is real and accelerating.

For builders, the practical question is: can you run your agentic coding workflows on an MIT-licensed model instead of paying per-token to Anthropic or OpenAI? Not today, for most people. A 754B MoE model requires serious inference infrastructure even with only 40B active parameters. But the direction is clear. In 12 months, smaller distillations of these models will run on the kind of hardware you already own.

What to do now: Don't switch your daily driver yet. Do add GLM-5.1 to your evaluation list for any self-hosted coding agent workflows. If you're building a product that depends on an LLM for code generation, start testing open-weight alternatives alongside your proprietary provider. The licensing freedom alone is worth the evaluation time.