GLM-5.3 is not out yet. As of July 20, 2026, Z.ai (Zhipu AI) has published no release note, model card, benchmark, or date for it, and the newest model the company ships is GLM-5.2, dated June 16. Everything circulating about GLM-5.3 is either a founder hint or community speculation. This piece separates the two: the verifiable release-timing math, the signals worth watching, what the model will probably change, and the model you can run today while you wait.
Is GLM-5.3 out yet?
No. The clearest evidence is Z.ai's own release-notes page, which lists GLM-5.2 (June 16, 2026) as the most recent model. There is no GLM-5.3 entry, no model card, and no benchmark table.
The z.ai homepage says the same thing in plainer terms: the assistant is "powered by GLM-5.2." So "GLM-5.3" is a community-used name for the expected next incremental release in the GLM-5 line, not a product you can call today. Treat any post claiming otherwise as unverified until it points to that release-notes page updating.
When will GLM-5.3 release? What the cadence says
Z.ai has not announced a date, so the honest estimate comes from its own shipping rhythm. The GLM-5 series has moved on a roughly two-month cadence:
- GLM-5: February 12, 2026
- GLM-5.1: April 7, 2026 (54 days later)
- GLM-5.2: June 16, 2026 (70 days later)
The first two bars are actual release gaps; the third is a projection (the 54-to-70-day midpoint), not a confirmed date.
Extrapolating that 54-to-70-day gap from the June 16 GLM-5.2 launch points to a mid-to-late August 2026 window (roughly Aug 9 to Aug 25). That aligns with community speculation on X, where accounts like @AndAtzx and @PrKingLondon have floated an August drop. Two intervals is a thin basis for a forecast, so treat this as a low-confidence range, not a countdown. Zhipu could slip it, skip a .3, or jump to a different version number.
The signals worth watching
Not every "GLM-5.3 soon" post carries the same weight. Here is how the current ones rank.
The founder's vision poll (the strongest official-adjacent signal)
On June 29, Zhipu co-founder Jie Tang ran a public developer poll asking whether the next GLM should include vision, and the response was near-unanimous yes (reported by TechTimes). This is community feedback, not a product announcement, and a poll about a wanted feature does not guarantee it ships in the next numbered release. But it is the closest thing to a roadmap hint from someone who sets the roadmap, and it puts multimodality clearly on the table.
Community chatter on GitHub and X
The GLM-5 GitHub repo already has open issues that name "GLM-5.3" in their titles. Read those carefully. At least one high-profile issue is a proposal from an outside researcher, not a Zhipu spec leak, and its body discusses GLM-5.2 internals rather than confirmed 5.3 features. It shows the community is building for 5.3, not that 5.3 has leaked. On X, the mentions are anticipation ("GLM 5.3 Soon!") rather than hands-on reports. As of this writing, no 5.3 weights, config files, or working API access have surfaced on Z.ai's official repos or in the public X posts I checked.
How the last version leaked
There is precedent for GLM releases surfacing early. GLM-5 was spotted before launch through GitHub pull requests and a zodiac-themed easter egg in the code, while Zhipu was mid-IPO. That is one prior signal, not a guaranteed pattern, but it is the reason repo activity is the first place worth checking for an early GLM-5.3 sighting.
What GLM-5.3 will likely change
Everything here is expectation, not confirmed spec. The table separates each claim from the signal behind it and how much weight it deserves.
| Expected change | Evidence behind it | Confidence |
|---|---|---|
| Native vision / multimodality | Founder's June 29 vision poll; existing GLM-5V multimodal line | Medium |
| Steadier long-context behavior | GLM-5.2 already ships 1M lossless context; .x releases tend to harden the prior version | Medium |
| Stronger agentic-coding reliability | Whole GLM-5 line is positioned for "agentic engineering"; known gap between benchmark scores and real-world robustness | Low-medium |
| Pricing in the current cheap bracket | GLM-5 open-weight tokens have run cheap; no confirmed 5.3 pricing either way | Low |
If GLM-5.3 ships without vision, that is the single biggest thing to fact-check against the poll signal.
What you can use today: GLM-5.2
Waiting for 5.3 doesn't mean waiting to build. GLM-5.2 is the current open-weight flagship. As a quick smoke test on July 20, I called the glm-5.2 model through an Anthropic-compatible relay with a "merge overlapping intervals" function plus a step-by-step trace on [[1,3],[2,6],[8,10],[15,18]]. It returned the correct output ([[1,6],[8,10],[15,18]]) on the first try in about 31 seconds, and the code noted it was avoiding mutating the input list. That is one informal call, not a benchmark, and a single interval-merge task says nothing about agentic or long-context performance. It is only a sanity check that the current model handles a routine coding prompt cleanly.
On the spec sheet, GLM-5.2 offers 1M-token lossless context and, per Z.ai's own release notes, open-source SOTA results on coding and long-horizon benchmarks. On cost, GLM-5 (the series baseline) has run around $0.66 per 1M tokens blended per Artificial Analysis, well under closed frontier models in the same class. That is the baseline GLM-5.3 will build on, which makes 5.2 a reasonable default to standardize on now rather than stall.
Whether to adopt 5.2 now or hold for 5.3 comes down to what you need:
| Use GLM-5.2 now if… | Wait for GLM-5.3 if… |
|---|---|
| You need stable, production-ready behavior today | You can pause a few weeks for a possible late-August drop |
| Your workload is text and code | Native vision/multimodality is a hard requirement |
| You want the lowest per-token cost in this class | You want the newest benchmark numbers before committing |
| You're building a pipeline you can re-test on upgrade | You're only experimenting and want bragging-rights latest |
The official path is Z.ai's own model API. If you would rather not manage a separate Z.ai account, GLM-5.2 is also reachable through Anthropic-compatible relay endpoints: see the GLM-5.2 API setup and how to run GLM-5.2 inside Claude Code. When 5.3 arrives, new GLM models have appeared on the same channels shortly after launch, but treat a version bump as a real change: re-run your regression tests before swapping model names, since context limits, tool-calling behavior, and pricing can shift between versions.
How to be first when GLM-5.3 drops
If catching the release early matters, watch these four places in order of reliability:
1. Z.ai release notes (docs.z.ai): the authoritative source, where the model card and date land first. 2. Hugging Face, zai-org: open weights publish here, often the same day. 3. GitHub, zai-org: PR and commit activity is where the last version leaked ahead of the announcement. 4. Jie Tang's social posts: the founder telegraphs direction, like the vision poll, before official channels update.
Skip the aggregator "leak" posts unless they link back to one of these. On brand-new Chinese model drops, X tends to lag the official channels, not lead them.
FAQ
How can I verify whether GLM-5.3 has launched?
Check the Z.ai release notes and the zai-org Hugging Face org. If neither shows a GLM-5.3 model card, it hasn't shipped, regardless of what aggregator posts claim. As of July 20, 2026, the newest listed model is GLM-5.2.
When will GLM-5.3 be released?
No date is confirmed. Based on Z.ai's two-month release cadence, a late-August 2026 window is a reasonable low-confidence estimate, not an announcement.
Will GLM-5.3 have vision?
Possibly. Co-founder Jie Tang's June 29 developer poll asked exactly this and the community answer was overwhelmingly yes, but a feature poll is not a confirmed spec.
Will GLM-5.3 be open-source?
Very likely. Every GLM-5 model so far has shipped with open weights on Hugging Face, and Z.ai markets the line as open-source SOTA. Expect the same for 5.3 unless the company signals otherwise.
How much will GLM-5.3 cost?
No pricing yet. GLM-5 has run around $0.66 per 1M tokens blended (Artificial Analysis), far cheaper than closed frontier models, so GLM-5.3 will most likely stay in that value bracket.