Tag
Databricks tested GLM-5.2, an open-source coding model, and found it competes with top closed models like Claude Opus 4.8 on real enterprise code tasks while being cheaper ($1.28/task vs $1.94/task). The evaluation also highlighted Pi, a harness that reduces costs by sending less context per turn.
A developer created a tool that chains a small local model with a larger coding model, automatically offloading VRAM between them to optimize memory usage.
Introduced MTP speculative decoding to the Ornith 35B coding model in FP8 precision, achieving approximately 18% faster inference with minimal extra VRAM.
Cohere Labs releases North Mini Code, a 30B parameter (3B active) open-source coding model under Apache 2.0, optimized for code generation and agentic tasks, capable of running locally with 20GB RAM via 4-bit quantization.
Kimi released the K2.7 Code model and its high-speed version, and announced API pricing. Compared to rival Mimo, it is more expensive and slower.
Open-source model builders have released Ornith-1.0-397B, a top-tier coding model, while larger models like Fable 5 and GPT-5.6 face restrictions.
Kog has open-sourced the Laneformer 2B model, a 2.3B parameter instruction-tuned coding model designed for high-speed decoding, achieving over 3,000 tokens per second by prioritizing latency from the architecture stage.
Jackrong releases Qwopus3.6-27B-Coder-Compat-MTP-GGUF, a GGUF quantization of the Qwopus3.6-27B-Coder model with an expanded chat template for better interoperability with tool-using runtimes and OpenAI-compatible agent frameworks.
GLM 5.2 is a frontier open-source coding model that performs near Claude Opus quality on coding tasks, with excellent tool calling, planning, and local deployment capabilities, at no cost.
The article discusses how JetBrains is quietly building an AI ecosystem for developers with tools like AI Assistant, their own coding model Mellum, and open-source AI projects, potentially positioning itself as a major AI winner.
Zhipu released the GLM 5.2 model, focusing on coding capabilities, open-source and supporting 1M context. Tests show it approaches Claude Opus 4.8 level in large engineering and coding tasks, but lacks multimodal capabilities and is limited by computational power, resulting in slower speed. The article also mentions Anthropic shutting down Fable 5 and Mythos 5 at the request of the U.S. Department of Commerce, highlighting the contrast between open-source and closed AI.
Zhipu AI announces GLM-5.2, their most capable open-source model with a 1M context window, positioning it as a foundation for complex agent applications and coding models, with immediate availability to GLM Coding Plan users and API next week.
Kimi released and open-sourced Kimi 2.7 Code, a coding model with improved performance, reduced reasoning tokens, and long-horizon coding abilities.
Cohere officially launches North Mini Code, a coding model, with weights available on Hugging Face and deployment support for vLLM and MLX.
Cohere released its first open-source coding model, North Mini Code Small, designed for efficient agentic performance and community input.
Cohere has released an early access coding model, BLS-Mini-Code-1.0, a 30B parameter model available on Hugging Face for testing.
A monokernel approach for LLM decoding on AMD MI300X GPUs achieves up to 3,300 output tokens/s per request without speculative decoding or quantization, using memory access patterns mapped to the die topology.
阿里巴巴发布了通义千问 3.7 Max，一款专为智能体时代设计的旗舰编码模型。该模型在长周期自主执行、前端生成和3D场景构建上表现突出，多项基准测试中与顶尖闭源模型持平甚至超越，是接近前沿的中国模型。
The user discusses their experience with Qwen 3.6 27B for local coding tasks and asks for recommendations for larger models (100B+) suitable for systems with 224GB of VRAM.
OpenAI releases GPT-5.2-Codex, an advanced agentic coding model optimized for complex software engineering tasks with improved long-horizon capabilities, Windows support, and cybersecurity features. The release includes comprehensive safety documentation through a system card outlining model and product-level mitigations.