In light of this incident, what would be a reasonable range of cybersecurity capabilities for models accessible to the general public, including the open-source community? In other words, how asymmetric should access to cybersecurity capabilities be?

Today marks exactly one month since GLM-5.2 launched, yet it already feels like a different world. “Closed” and “open” models no longer represent a capability gap or an East–West divide; the distinction now simply comes down to whether a model is open.

Finding a sustainable path for products like Kilo Code is never easy, and we’ve spent a lot of time together exploring what that path could look like. Really happy to see Kilo reach this important milestone.

Our Single-rollout Asynchronous Optimization (SAO), is able to train stably for one thousand steps and consistently outperform GRPO and its variants on agentic coding and reasoning benchmarks, such as SWE-Bench Verified, BeyondAIME, and IMOAnswerBench.

Great to see GLM-5.2 neck-and-neck with Opus 4.8 on AA and Harvey’s new legal agent benchmark, an area that doesn’t get discussed enough. Curious what matters most in real legal workflows: deeper legal knowledge, stronger reasoning, better retrieval, or tighter review loops?

slime now adds --release-train, pushing the inference system during agentic RL training to a new limit. In colocated RL training, we want SGLang to use as much room as possible for inference-side optimizations such as HiCache, instead of being constrained by offloaded Megatron training processes. --release-train makes this possible by releasing the Megatron training process during rollout and reloading it for each training round. This gives SGLang more configuration headroom in colocated RL workloads. PR:

GLM-5.2 is now selectable in Claude Code via Hugging Face🤗 Inference Providers + hf-claude. Open models are becoming easier to plug directly into real developer workflows. 😀

I was tracking down GLM 5.2 accuracy issue on NVFP4 (endless !!!!) and proposed the following fix, it works for me but if it doesn't help with your setting, let me know your running config! Model:

Looks like 'vision' is taking over the comment section. If you want to discuss something other than vision & token efficiency, such as a particular specialized capability, feel free to share your thoughts.

The RL framework behind GLM-5.2 is fully open source. The full post-training of GLM-5.2 ran on it in about two days. The same stack sits behind the entire GLM series, from 4.5 to 5.1. It is called slime, and it is built around one idea. Keep a single RL kernel, and push all the variety into data generation. Let me explain what that means. Every RL run has two halves. One generates experience, where the model produces responses and something scores them. The other learns from it by updating weights. The learning half is mechanical. It reads samples, computes a loss, and steps the optimizer, the same way whether the model solves equations or drives a browser. What changes between tasks is generation. A math run answers in a single turn and grades the result. An agent run loops through tool calls, reads results, and only then earns a reward. slime draws the line right there. The learning half stays fixed as one kernel, and everything that differs becomes a new way to generate data. Under the hood, it wires Megatron for training to SGLang for rollout, with a Data Buffer between them that owns prompts, custom data, and generation. Most RL stacks grow into a pile of disconnected trainers, rollout services, and agent frameworks. slime refuses that. Multi-turn tool use, sandbox interaction, environment feedback, and verifier rewards all enter as data generation, not as forks of the loop. So an agentic workload runs on the same loop a math run uses, and the kernel never changes. A few things follow. → It is battle-tested. The loop is validated by shipping real GLM models, and it also supports Qwen3, DeepSeek V3, and Llama 3. → Correctness comes first. RL bugs are silent, so slime keeps the dataflow explicit and treats CI, reproducibility, and fault tolerance as real engineering. The proof is the ecosystem on top of it. Dressage, Miles, vime, Relax, OpenClaw-RL, P1, and TritonForge all build on slime without touching the core loop. The lesson is not that RL needs a bigger framework. It is that the variety belongs in data generation, and the training loop should stay small enough to trust. GitHub repo: (don't forget to star 🌟) Since we're talking about RL, I wrote a full breakdown on fine-tuning LLMs with RL in 2026. Including how to skip manual reward engineering with automatic LLM-graded rewards. The article is quoted below.