Zhipu's GLM-5.3-Highspeed Spotted in Grayscale at 490 Tokens/s
量子位 · wechat · 2026-10-12
A user reports that Zhipu's new GLM-5.3-Highspeed is rolling out randomly to Zcode 3.15.1 subscribers, with a leaked speed of 490 tokens/s — versus 200 tokens/s for GLM-5.3-FlashX released last month and 60 tokens/s for GLM-5.3-Flash on Z.ai's API.
Pi's model catalog already lists it: 1M-token context window, 131,072 max output tokens, and low/high/max thinking-intensity tiers, matching the standard GLM-5.3 config. Reddit users speculate it's the anonymous Omen Alpha on OpenCode (widely believed GLM-related); hands-on impressions say it's fast, thinks long, and performs well, though its true identity is unconfirmed. Zhipu shipped a 400 tokens/s GLM-5.1-highspeed in May; if 490 TPS holds, this would set a new API speed record. Unofficial, still in grayscale rollout.
More from Models
- Commenter: Anthropic keeping old models alive exposes its cost excuse — repligate · 2026-10-12
- The model swallows the harness: agent frameworks' logic migrating into models — khademinori · 2026-10-12
- Claude Code Users Report Fast Mode Locked Despite Plenty of Remaining Credits — RileyRalmuto · 2026-10-12
- GPT-Live users report constant "give me a minute" delays in Instant mode — Key-Astronaut2407 · 2026-10-12
- OpenAI's new Decisions API fails a simple coin-flip calibration test — JnBrymn · 2026-10-12
- Unverified leak claims Anthropic's next Claude is being tested in Claude Code, billed as SOTA — Eric520CC · 2026-10-12