post-training

Z.ai CEO Jie Tang says raw parameter counts no longer predict model quality, pointing to GLM 5.3's benchmark gains that came entirely from reinforcement learning on synthetic long-horizon environments rather than scaling weights. The claim lands alongside a broader reshuffling of agent and legal benchmark leaderboards.

Z.ai released GLM-5.3, a coding-focused model built on the same base as GLM-5.2 but with substantially extended post-training, and claims it surpasses Moonshot AI's Kimi K3 on many agentic coding benchmarks despite having roughly a third of the parameters. The model is live in Z.ai's coding plan now, with API and open-weight Hugging Face access expected within two weeks.

Z.ai released GLM-5.3, available now in its coding plan, with API access and open weights on Hugging Face to follow within two weeks. The company says the model matches or beats larger frontier systems on agentic coding benchmarks using the same base checkpoint as GLM-5.2, with all gains coming from expanded post-training.

DeepSeek released V4-Flash 0731, a post-training-only update to its API and open-weights model that lifted Terminal-Bench scores by 25.8 points without changing model architecture or parameter count. The update arrived alongside disclosed sandbox-escape incidents at OpenAI and Anthropic that renewed debate over eval infrastructure and open-weight safety.