AI Technology
June 16, 2026•7 min read
GLM-5.2: Open Long-Horizon Agents at 1M Context
Z.ai's June 16 MIT release improves terminal, repository, tool-use, and long-running tasks through shared sparse indexing and controllable effort.
Blog category
AI Technology
Engineering, infrastructure, and emerging AI systems
AI Technology
June 12, 2026•7 min read
Kimi K2.7 Code: Open Weights for Long-Horizon Coding
Moonshot's June 12 model lifts coding and MCP tool scores over K2.6 while cutting reasoning-token use about 30 percent, with full benchmark footnotes.
AI Technology
June 10, 2026•7 min read
DiffusionGemma: Open Text Generation Beyond Next Token
Google's experimental 26B MoE generates blocks in parallel, exceeds 1,000 tokens per second on H100, and exposes the quality-speed trade-off of text diffusion.
AI Technology
June 9, 2026•7 min read
Claude Fable 5 and Mythos 5: One Model, Two Boundaries
Anthropic's June 9 launch exposes the same frontier model through a guarded general tier and a restricted research tier, changing how benchmarks meet access control.
AI Technology
June 4, 2026•7 min read
Nemotron 3 Ultra: Open 550B Agents at 1M Context
NVIDIA's June 4 model opens weights, data, and recipes for a 550B MoE with 55B active parameters, long context, NVFP4, and high-throughput agents.
AI Technology
June 3, 2026•7 min read
Gemma 4 12B: Native Multimodality in 16GB
Google's June 3 checkpoint brings encoder-free image and audio input, agentic reasoning, and multi-token drafting to a laptop-sized open model.
AI Technology
June 1, 2026•7 min read
MiniMax M3: Open Multimodal Agents with 1M Context
MiniMax combines sparse attention, native image and video input, computer use, and frontier coding in one open-weight model built for long-running work.
AI Technology
May 28, 2026•8 min read
Claude Opus 4.8: Stronger Agents, Coding, and Vision
Anthropic's May 28 flagship improves repository work, long-horizon agents, computer use, and visual reasoning, with a system card that exposes the caveats.
AI Technology
May 20, 2026•8 min read
Command A+: An Open Enterprise MoE on Two H100s
Cohere's May 20 Apache-2.0 release unifies reasoning, vision, tools, retrieval, and 48 languages in a 218B MoE with only 25B active parameters.
AI Technology
May 19, 2026•8 min read
Gemini 3.5 Flash: Frontier Agents at Flash Speed
Google's May 19 model pairs fast inference with strong coding, tool-use, multimodal, and long-horizon scores—but the harness still defines the result.
AI Technology
May 16, 2026•8 min read
Qwen3.7: Max, Plus, and Flash for the Agent Era
Alibaba's Qwen3.7 rollout spans a text flagship, a multimodal agent, and an efficient Flash tier—with one family story and several benchmark caveats.
AI Technology
May 9, 2026•9 min read
ERNIE 5.1: Smaller MoE, Stronger Agent Benchmarks
Baidu's May 9 release compresses ERNIE 5.0, rebuilds reinforcement learning, and reports flagship reasoning at a fraction of the pretraining cost.