GLM-6
By joining you agree to the ArtificialWatch Terms of Service and Privacy Policy. You'll get the alerts you picked; unsubscribe in one click. 18+.
Not out yet. Zhipu AI hasn't released GLM-6. Zhipu's Z.ai shipped a new GLM-5.x about every two months this year: GLM-5.2 on June 16, then GLM-5.3 on August 14 and GLM-5.3 Flash on August 26, both read straight off the catalogs by our own sweep. On August 31 founder Tang Jie named GLM-6.0 as the next base model, a “full self-training” system, but Zhipu has given no date, price or model id. As of November 7 it has been 85 days since GLM-5.3, the longest gap between GLM versions this year. In late October Zhipu told its paying GLM Coding Plan users a new model is coming, without naming it or giving a date. We sweep the API every 60 seconds. The second GLM-6 answers, ArtificialWatch fires.
Leaks & news, newest first.
85 days since GLM-5.3, and Zhipu has told paying users a new model is coming
As of Nov 7 it has been 85 days since GLM-5.3, against gaps of 54, 70 and 59 days between the last three versions. Read on Nov 7: Z.ai’s English and Chinese release notes still end at GLM-5.3 Flash (Aug 26), its newest Hugging Face repo dates from Aug 25, and OpenRouter’s newest Z.ai id is still the legal-domain GLM-5.3 FlashX (Nov 2). Neither OpenRouter nor models.dev lists a glm-5.4, glm-5.5 or glm-6 id. The one word from Zhipu itself came in a compensation notice it sent GLM Coding Plan subscribers in late October, after some users reported weaker answers: it granted three quota resets and promised another one “after the new model is released”, per QbitAI (Oct 23). The notice names no model and no date, so whether that model is GLM-5.5 or GLM-6 is the outlets’ guess, not Zhipu’s. Polymarket’s next Z.ai GLM (5.4+) market is unchanged at 56% by Nov 30 and 72% by Dec 31. via QbitAI · Z.ai docs · Hugging Face · OpenRouter · models.dev · Polymarket
82 days since GLM-5.3, and no GLM-6 id anywhere
Z.ai shipped GLM-5.1, 5.2 and 5.3 at gaps of 54, 70 and 59 days. As of Nov 4 it has gone 82 days since GLM-5.3 without a new version, the longest gap this year. Z.ai’s release notes still end at GLM-5.3 Flash (Aug 26), its newest Hugging Face repos date from Aug 25, and no glm-6 id is in OpenRouter’s catalog. Polymarket’s “next Z.ai GLM (5.4+)” market, which any GLM-5.4 or later would settle, GLM-6 included, last traded at 56% by Nov 30 and 72% by Dec 31 (thin books, so we quote the last trade, as Polymarket does, not the midpoint); its Oct 31 line resolved no. via Z.ai docs · Hugging Face · OpenRouter · Polymarket
A legal-domain GLM-5.3 FlashX lands on OpenRouter
OpenRouter added z-ai/glm-5.3-flashx-legal, which it describes as Z.ai’s legal-domain variant of GLM-5.3-FlashX, with extra legal mid-training and post-training. It extends the 5.3 family; it is not a new GLM version. via OpenRouter
Tang Jie: self-improvement is “the main research problem” for AGI
At an event in Shanghai, Z.ai co-founder Tang Jie called recursive self-improvement “the main research problem for [achieving] AGI” and said it has to be done safely, per the South China Morning Post. It is the direction he set for GLM-6.0 in August. via South China Morning Post
Tang Jie names GLM-6.0: a “full self-training” model
At Zhipu’s first-half results briefing, founder Tang Jie defined GLM-6.0 as Full Self-Training, which he likened to what the West calls recursive self-improvement: the model taking part in its own pre-training, mid-training and post-training. The hard part, he said, is the model judging for itself when to stop and when to correct itself. Zhipu also said its next-generation base model is already in training, aiming for a larger effective size, longer native context and native multimodality while keeping active parameters in check. No date, price or model id. via Guandian · National Business Daily
GLM-5.3 weights go public, two weeks after launch
Z.ai published GLM-5.3 on Hugging Face under its own GLM-5.3 license, inside the two-week window it gave at launch. The 5.x point releases have not paused yet. via Z.ai · Hugging Face
GLM-5.3 Flash ships with MIT weights
Z.ai released GLM-5.3 Flash, a 320B-parameter MoE with 18B active and native visual input, and published its weights on Hugging Face under the MIT license. via Z.ai docs · Hugging Face
GLM-5.3: same base as 5.2, gains from post-training
Z.ai launched GLM-5.3 on the same base model as GLM-5.2, crediting its gains to scaled-up post-training for coding and cyber defense. It went to the API and the GLM Coding Plan first, with weights promised within two weeks. via Z.ai · SiliconANGLE
Watching — the six is unannounced
No date, no teaser, no benchmark leaks. GLM-6 is on the board because of who's behind it, not what's been said: Zhipu has out-shipped nearly every CN lab this year.
GLM-5.2 ships — the base 5.3 would build on
Z.ai pushed GLM-5.2 mid-June, the latest in a steady run of 5.x point releases. Each one raises the floor the eventual GLM-6 has to clear. via Z.ai · llm-stats
The tell to watch: point releases stop, then the number jumps
The GLM line historically goes quiet on point releases a few weeks before a major version. When the 5.x drumbeat pauses, that's the signal — and the API sweep catches the rest.
The official word.
Zhipu has named GLM-6.0 but announced no date, price or model id. Founder Tang Jie described it on the Aug 31 results call as a full self-training model, and the Aug 26 GLM-5.3 Flash release is still the newest entry in Z.ai’s release notes. Zhipu’s late-October notice to Coding Plan subscribers says only that a new model is coming. When the six answers, it lands on this page and your alert channels the same minute.
ArtificialWatch wire desk · Nov 7, 2026
Until GLM-6 ships: GLM-5.3.
Which model is current until GLM-6 ships?
The current model in this line is GLM-5.3 from Zhipu AI, live since August 14, 2026.
$1.40 / $4.40 per 1M tokens on Z.AI's API.
Hear it the second GLM-6 answers.
We sweep the public catalogs (OpenRouter, models.dev) and Z.ai's own docs every 60 seconds — that is how GLM-5.3 was caught. The moment GLM-6 returns a real answer, a text or a phone call fires on paid tiers; free email + Chrome push follow about 15 minutes later.
By joining you agree to the ArtificialWatch Terms of Service and Privacy Policy. You'll get the alerts you picked; unsubscribe in one click. 18+.