A JPMorgan research note, relayed by and CGTN in late June 2026, put a trillion-parameter open-weight model from Zhipu AI on the calendar for August. The note described the model as a "next key milestone in Chinese frontier AI" and cited a parameter count exceeding 1 trillion. Community leaks that followed in mid-July added a 1-million-token context window, an MIT open-weight license, and a version jump from GLM-5.2 straight to 5.5 — skipping 5.3 and 5.4 entirely.

As of early August 2026, Zhipu AI (now branding as Z.ai) has not published a model card, a blog post, a Hugging Face repository, or an API endpoint for GLM-5.5. The company's official channels still promote GLM-5.2, released June 13. The August date, the parameter figure, the license assumption, and even the 5.5 naming all rest on analyst forecasts and secondary reporting — not on a vendor announcement.

Confirmed

  • JPMorgan analysts forecast a new Zhipu model for August 2026; carried the note on June 25, and CGTN republished it on June 30 while covering the GLM-5.2 launch.
  • The research note describes a model with more than 1 trillion total parameters and frames it as a milestone for Chinese frontier AI.
  • Zhipu founder Tang Jie publicly replied "epic plus" (史诗级plus) on July 20 to a user asking whether GLM still had hope after Qwen and Kimi upgrades — confirming a meaningful update is in preparation, but not naming it GLM-5.5 or confirming August.
  • GLM-5.2, the current flagship, is a 744-billion-parameter mixture-of-experts model with ~40 billion active parameters, a 1-million-token context window, and MIT-licensed open weights released June 13, 2026.
  • Zhipu has publicly committed to releasing an open-weight model at "Fable/Mythos" frontier tier before the end of 2026; GLM-5.5 is widely treated as the next step toward that goal.

Analysis

The evidence chain is real but thin: an analyst forecast picked up by wire services, amplified by state media, and then elaborated by community leaks. That chain is stronger than an anonymous social post, but it still terminates at a research note — not a Z.ai release note. The parameter count (>1T), the context window (1M tokens carried from GLM-5.2), the license (MIT by precedent), and the version number (5.5 vs. 5.3 vs. 6) all share the same status: unconfirmed extrapolation.

The two-month cadence (February, April, June) makes an August release plausible on rhythm alone. But cadence is not commitment. Zhipu has shipped point releases (5.1) and major releases (5.2) on that schedule; skipping two version numbers suggests the team believes the delta is large enough that 5.3 would undersell it. That signal aligns with Tang Jie's "epic plus" framing and with analyst Teortaxes' observation that GLM-5.5 "might well be a new and larger pretrain" targeting the Fable/Mythos tier.

Hardware planners should note: even a >1T total MoE model can remain cheap to serve if active parameters stay near GLM-5.2's ~40B. But without a confirmed active-count, VRAM estimates are guesswork. The same applies to pricing — GLM-5.2's API runs roughly $1.40 in / $4.40 out per million tokens, but a successor can change license, modalities, and access paths.

Unknown

  • Exact release date — "August 2026" is a JPMorgan projection, not a Zhipu commitment.
  • Architecture — whether GLM-5.5 is a scaled-up MoE, a dense model, or a new design; active parameter count is unknown.
  • Version naming — developer discussions use GLM-5.3 for a possible point release; analyst coverage uses GLM-5.5; some Chinese reporting speculates GLM-6. None is confirmed.
  • Multimodal support — vision, audio, or other modalities have not been addressed in any confirmed source.
  • Benchmarks — no scores, no model card, no third-party evaluations exist.
  • API availability — release does not mean immediate endpoint; GLM-5.2's route took days to stabilize across providers.

Our take

The August rumor cluster is credible enough to watch — not to treat as a product. Until Zhipu publishes a model card, repo, or API entry, GLM-5.5 remains an analyst signal layered on top of a real GLM-5.2 release. The useful distinction for readers is simple: ship and evaluate what is callable today; keep the trillion-parameter story in the Unknown column.

Series: 1. GLM-5.2: What Actually Shipped and Why It Matters · 2. The 2026 Roadmap: Zhipu's Release Cadence and Strategy · 3. GLM-5.5 Rumors vs. Official Silence: Parsing the August 2026 Reports · 4. Developer's Guide: What to Treat as Confirmed vs. Unknown in the GLM Ecosystem · 5. GLM-5.3 Ships: Post-Training Coding and Emergent Cyber Defense · Zhipu GLM

Sources