GLM-5.3 (Inference.net)

Z.ai launched GLM-5.3 on August 14, 2026, with substantial gains in long-horizon coding and a notable jump in cybersecurity capabilities. VentureBeat reports GLM-5.3 was available initially only through Z.ai's GLM Coding Plan and ZCode environment, with API access and open weights delayed pending safety evaluation and hardening. Z.ai plans to release open weights roughly two weeks after launch. The release is positioned as a stress test of how far a frontier-scale base model can be pushed via post-training alone. GLM-5.3 builds on the 743-billion-parameter-scale base behind GLM-5.2 rather than replacing it; Z.ai expanded its post-training system using more environments and additional reinforcement-learning compute. According to VentureBeat, Z.ai stated "Scaling post-training is all we did for GLM-5.3." Cybersecurity capabilities reportedly improved faster than anticipated as training scaled, producing unusual safety considerations for an open-model developer and prompting a reported "trusted access" approach for sensitive functionality.

Z.ai released GLM-5.3 on August 14, 2026, achieving frontier-level coding benchmark performance from a roughly 750-billion-parameter base — about a third the size of Moonshot AI's Kimi K3. The Interconnects analysis notes GLM-5.3 surpasses Kimi K3 on many benchmarks and matches or exceeds Claude Fable 5 and GPT-5.6-Sol on some, positioning it near the agentic coding frontier. Availability was limited to Z.ai's coding plan at launch, with API and Hugging Face open weights expected to follow. GLM-5.3 reuses the same base model as GLM-5.2, with all gains coming from expanded post-training — more environments, diverse tasks, and additional reinforcement-learning compute rather than a fresh pretraining cycle. Interconnects frames this as evidence of Z.ai's particular strength in post-training relative to Kimi's pretraining focus. The piece frames the release as a test of how far a frontier base model can be pushed without another expensive pretraining round.