China’s Zhipu AI released GLM 5.2 on June 13, 2026 — a coding-first large language model with a 1 million token context window, up to 131,072 output tokens per response, and an MIT-licensed open-weights release arriving this week. The timing is striking: Anthropic’s Fable 5 and Mythos 5 were shut down globally by a US government export directive just one day earlier, leaving developers who relied on Fable 5 for long-context coding work looking for alternatives.
GLM 5.2 is now live on all four tiers of Zhipu’s GLM Coding Plan — Lite, Pro, Max, and Team — through the Z.ai platform. Open weights and a standalone API are expected during the week of June 16.
What’s New in GLM 5.2
GLM 5.2 is built on the same 744-billion-parameter Mixture-of-Experts (MoE) architecture as its predecessor GLM-5. The key upgrades are:
- 1M token context window — labeled
glm-5.2[1m]in the API, matching the context window Fable 5 launched with before its ban - 131,072 output tokens per response — roughly five times the output window of GLM-5.1’s 200,000-token input window
- Dual thinking modes — High (efficient) and Max (recommended for complex coding tasks)
- MIT license — open weights with permissive commercial use, no export controls
The model is designed specifically for coding. Zhipu lists compatibility out of the box with Claude Code, Cline, OpenCode, Roo Code, Goose, Crush, OpenClaw, and Kilo Code — a deliberate targeting of the same developer tooling ecosystem that Anthropic’s models support.
One Caveat: No Independent Benchmarks Yet
Zhipu has not published independent benchmark results at launch. There are no HumanEval, SWE-Bench, or MMLU scores to compare against GPT-5.5, Claude Opus 4.8, or Gemini 3.5. The company describes performance as “powerful coding” and “strong long-horizon” reasoning — claims that cannot be independently verified until the open weights ship and the community runs evaluations.
Leaderboards like Aider.chat’s Coding Leaderboard and BigCodeBench will be the right places to watch once weights are public.
The Geopolitical Context
GLM 5.2’s launch lands at an unusual moment. The same week the US government used export control powers to shut down Anthropic’s two most capable models from every foreign national globally, China’s Zhipu released a frontier-class coding model under an MIT license with no geographic restrictions.
The MoE architecture means GLM 5.2’s 744 billion total parameters activate only a fraction per token, keeping inference costs competitive. Once the quantized open weights ship, high-end consumer hardware — GPU setups like those already running 27B models at 80 tokens per second — should be able to run smaller quantized versions locally.
For enterprise teams and independent developers outside the US who lost access to Fable 5, GLM 5.2 is the most direct open-source substitute currently available — provided the performance claims hold up under independent evaluation.
Our Take
GLM 5.2 is well-timed and well-specced, but the missing benchmarks are a real gap. Zhipu’s prior models — GLM-4 and GLM-5 — delivered on their architectural promises once tested. If GLM 5.2’s weights arrive this week and the community validates coding performance anywhere near Fable 5’s level, this becomes one of the more significant open-source releases of 2026. Check back when independent evals land.
Frequently Asked Questions
Is GLM 5.2 available now?
The cloud API through Z.ai’s GLM Coding Plan is live now. Open weights and a standalone API are expected during the week of June 16, 2026.
How does GLM 5.2 compare to Claude Fable 5?
Both feature a 1M token context window. Fable 5 is currently unavailable globally due to a US government export directive. No independent benchmarks comparing the two have been published yet — Zhipu has not released evaluation scores at launch.
Can I run GLM 5.2 locally?
Not yet. Open weights are arriving the week of June 16. The 744B MoE model at full precision requires significant GPU memory, but quantized versions (Q4/Q5) should run on high-end consumer hardware once available.
Is GLM 5.2 subject to US export controls?
GLM 5.2 is a Chinese open-source model released under an MIT license. It is not subject to US export control directives and is available to users globally without geographic restrictions.
