In coding AI, an open-weight model has pulled up right behind the best closed model. GLM-5.2, released by China's Zhipu (Z.ai) on June 13, 2026, scored within about 1~4% of Claude Opus 4.8 on FrontierSWE and Terminal-Bench, and its weights ship under the MIT license. Its price is about one-sixth of GPT-5.5. The top open-source coding model is thinning the moat around closed models. ASAP summarizes the result from the primary announcement.

It became the top open-source coding model

GLM-5.2 is the top open-source model for coding, ranking first among open-weight systems on public benchmarks. On authoritative benchmarks such as FrontierSWE and Terminal-Bench 2.1, GLM-5.2 came within about 1~4% of the best closed model, Claude Opus 4.8. On Code Arena, a blind evaluation with a million users, it placed first among all available models.

It released the weights under the MIT license

Zhipu opened GLM-5.2 in full on June 13, 2026 and published the weights under the MIT license. Anyone can download the model, self-host it, and use it commercially without restriction. Top-tier coding performance that used to live only behind a closed API has moved into open weights.

A genuinely usable 1M context

GLM-5.2 supports a 1 million token context and up to 128K tokens of output. Zhipu defines GLM-5.2 as a base model for agentic engineering, designed to handle long codebases and multi-step tasks in one pass. It also offers two reasoning modes with adjustable thinking effort.

The cost is about one-sixth of GPT-5.5

GLM-5.2 is priced at roughly $1.4 per million input tokens. For the same coding work, it reportedly costs about one-sixth of GPT-5.5. When performance is similar but price drops sharply, the unit economics of coding workloads change.

The caveats: self-reported scores and data risk

GLM-5.2's benchmark numbers are still Zhipu's own claims and have not been independently verified by a neutral harness. Using it via API also raises questions about China-based data handling. For sensitive code, self-hosting the open weights is the safer path.

What it means: the moat around coding models is thinning

GLM-5.2 shows that the lead of closed models in coding is narrowing fast. When an open-weight model lands within 1~4% of the best at one-sixth the price, the case for always paying for the most expensive model weakens. Picking a model per task, and using a cheaper one for verifiable coding, is what decides cost.

Wrap-up

GLM-5.2 is a case of an open-weight coding model closing the gap with the best closed model to 1~4%. Its MIT-license release, 1M context, and roughly one-sixth-of-GPT-5.5 price are the evidence. But self-reported scores and data risk remain, so weigh verification and self-hosting before adopting it.

Source: ASAP summary of Zhipu (Z.ai)'s GLM-5.2 release (June 13, 2026; MIT license, 1M context and 128K output, within about 1~4% of Claude Opus 4.8 on FrontierSWE and Terminal-Bench, about $1.4 per million input tokens) and related reporting.