GLM-5.2 release date: June 2026, and it is callable today.

Quick answer: the verified timeline.

GLM-5.2's release happened in two steps. On June 13, 2026, Z.ai shipped the model to subscribers of its GLM Coding Plan, the quota-based plan aimed at coding-agent users. On June 16, 2026, the date Z.ai's own release notes carry, the public release followed: MIT-licensed open weights on Hugging Face and standalone per-token API access. Independent coverage, including Simon Willison's June 17 write-up, matches that two-step sequence. So unlike most release-date searches, this one has a closed answer: the model is out, the weights are public, and the API id is stable. The open questions are what shipped exactly, and what is still unconfirmed about where the family goes next; both are covered below.

The GLM-5 family history: three releases in four months.

Z.ai's release notes date the family precisely, and the cadence is striking: three major releases between February and June 2026. GLM-5 arrived on February 12, 2026, positioned for complex system engineering and long-range agent tasks, and notable for integrating DeepSeek Sparse Attention for better token efficiency at long context. GLM-5.1 followed on April 7, 2026, with the autonomy claim that defined its launch: runs of up to 8 hours on a single task, with improvements in planning, sustained execution, bug fixing, and strategy iteration. GLM-5.2 closed the sequence on June 16, 2026, claiming open-source state of the art on coding and long-horizon benchmarks. A roughly two-month interval between releases is not a promise of the next date, but it is the observable pattern, and all three models remain served side by side rather than replacing each other, so nothing about a future 5.3 changes what you can build on today.

What shipped on June 16: verified vs reported.

One early observation from independent testing is worth carrying into cost planning: GLM-5.2 spends more output tokens per task than GLM-5.1 on the same benchmark (about 43K vs 26K in Simon Willison's measurement). Reasoning tokens bill as output, so the newest model is both the strongest and the heaviest of the family per task.

• Verified from Z.ai docs: model id glm-5.2, a 1M-token context described as lossless, up to 128K output tokens, a thinking mode with toggle and effort settings, tool calling, streaming, and structured JSON output.
• Verified from the public release: MIT license, weights on Hugging Face at roughly 1.5TB.
• Reported by launch coverage, treat as approximate: a Mixture-of-Experts around 750B total parameters with about 40B active per token; published counts differ slightly between 744B and 753B across sources.
• Reported by launch coverage, single-sourced: a SWE-bench Pro score of 62.1 against 58.6 for GPT-5.5, and a 28.5T-token training corpus. Neither figure appears in the vendor docs pages we checked, so weigh them accordingly.

What is not confirmed.

No GLM-5.3 or GLM-6 has been announced, and Z.ai has published no roadmap dates beyond the June release; anything circulating about the next version is inference from the family's two-month cadence, not vendor information. We could not confirm the exact parameter count from vendor documentation, which is why this page carries it as a range from launch coverage. And benchmark standings move quickly in 2026; the SWE-bench Pro figure above describes launch week, not necessarily today's leaderboard. The practical hedge against all of this is the same: integrate against the exact id glm-5.2 behind an endpoint that serves multiple families, and a future release, whether from Z.ai or a competitor, becomes a one-string migration you can A/B rather than an event you wait on.

How to call GLM-5.2 today.

The released model is addressable now through any OpenAI-compatible client: base URL https://api.apisrouter.com/v1, model glm-5.2, one key that also covers glm-5.1 and glm-5 for downshifting volume work. No Z.ai account or coding-plan subscription is required for API access through the gateway. Note one serving detail: the glm-5.2 catalog entry here lists a 200K context window, so budget prompts against 200K as called from this endpoint, whatever the upstream 1M figure supports.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APISROUTER_API_KEY"],
    base_url="https://api.apisrouter.com/v1",
)

resp = client.chat.completions.create(
    model="glm-5.2",   # released June 16, 2026
    messages=[{"role": "user", "content": "What changed between GLM-5.1 and GLM-5.2?"}],
    max_tokens=2000,
)
print(resp.choices[0].message.content)

Migrating from glm-5.1 or glm-5: what to check first.

The whole migration is an afternoon precisely because the family shipped as separate stable ids behind one request shape. That is also the argument for pinning exact ids in production configs: February code that said glm-5 still runs glm-5, and upgrades happen on your schedule with an eval in front of them.

• Quality is the easy half: rerun your existing evaluation set with the model string swapped. The request shape is identical across the family, so the diff is one line and the comparison is apples to apples.
• Output volume is the surprise half: 5.2 thinks longer than 5.1 on the same tasks by early independent measurements, and reasoning tokens bill as output. Compare completion_tokens per task across the two ids before comparing per-token rates.
• Price steps up at each rung: glm-5, glm-5.1, and glm-5.2 carry three different rates, so a workload that passes evals on a lower rung has no business on a higher one. The catalog table below has all three.
• max_tokens budgets need rechecking: a cap tuned for 5.1's reasoning volume can truncate 5.2 mid-thought on the same prompts. Raise the budget for the eval run, then tighten against observed usage.
• Nothing forces the move: all three ids stay served side by side, and mixed routing, 5.2 for the hard steps, older siblings for volume, is a supported steady state rather than a transition phase.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Why release dates for this family keep confusing people.

Three patterns generate most of the noise. First, staged access: June 13 and June 16 are both correct dates for different audiences, coding-plan subscribers versus everyone, and coverage written between the two reads as contradictory. Second, platform lag: a model can be live on the vendor API for days before third-party platforms list it, so "is it out?" genuinely has different answers per endpoint on launch week. Third, family naming: GLM-5, 5.1, and 5.2 are separate served models, not rolling updates to one id, so a search for "GLM-5 release" surfaces February news about a model that has since been superseded twice. The reliable check beats all of it: list the models your endpoint serves and look for the exact id. If glm-5.2 appears in /v1/models, it is released for you; everything else is press-cycle detail.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | grep glm-5.2

FAQ