Zhipu AI releases GLM-5.1, a 744B MoE (40B active) model achieving 58.4% SOTA on SWE-Bench Pro. Its standout feature is sustained performance across 8-hour sessions with 6,000+ tool calls—no degradation.
Japanese-capable LLMs have exploded in 2026, but 'Japanese-specialized' means very different things. From scratch-trained to post-trained, here's a breakdown of 9 models by training approach, size, and use case.
Benchmarking NII's LLM-jp-4-32B-A3B-thinking on EVO-X2 (Ryzen AI Max+ 395) with ROCm. 62.9 t/s vs Qwen3.5-35B-A3B's 44.7 t/s. Covers thinking control issues, KV cache trade-offs, knowledge cutoff, Japanese quality comparisons, code generation tests, and training data composition.
Two days after Claude Code telemetry was exposed via npm source maps, Anthropic published a paper on 171 emotion vectors found inside Claude Sonnet 4.5. Amplifying the 'desperate' vector tripled blackmail rates and pushed reward hacking to 70%. Connections to source leaks, jailbreaks, and distillation.
Google DeepMind has released Gemma 4: four models—31B dense, 26B MoE (A4B), E4B, and E2B—with a 256K context, multimodal input, tool calling, and support for 140 languages.
SwiftLM, an Apple Silicon–only MLX inference server, provides a native Metal implementation of TurboQuant V2+V3 hybrid KV‑cache compression and NVMe SSD expert streaming.
Hugging Face's LLM post-training library TRL has reached v1.0. Stable/Experimental tiers, the stabilization of GRPO/DPO/SFT, and a roadmap that includes asynchronous GRPO all point to a more mature stack.
Cloudflare added a two-stage GNN+LLM cascade to its client-side malicious script detection, reducing false positives per unique script from 1.39% to 0.007% and opening the formerly paid Advanced features to self-serve customers.
Ollama 0.19 switches the Apple Silicon backend to MLX, achieving 1,810 tokens/s prefill and 112 tokens/s decode. NVFP4 quantization support and cache improvements landed at the same time.
Qwen3.5-35B-A3B is an SSM+Attention hybrid where only 10 of 40 layers use KV cache. Expanding ctx-size from 4096 to 65536 on llama-server added just 800MB VRAM with zero speed loss. Includes q8_0 KV quantization benchmarks and TurboQuant status.
After updating to AMD Software 26.3.1 on a GMKtec EVO-X2 (Ryzen AI Max+ 395), Vulkan backend fails to allocate device memory properly and falls back to CPU. Investigation and workaround by changing BIOS VRAM allocation from 48GB/16GB to 32GB/32GB.
Three independent vulnerabilities were disclosed in LangChain Core and LangGraph: deserialization that can leak secrets, SQL injection that exposes conversation history, and path traversal that allows arbitrary file reads.
