coding

Mistral has released Mistral Large 4 ('Le Chonk'), a one-trillion-parameter open-weight model it claims is the most capable open-weight model outside of China, optimized for coding, cyberdefense, and domain-specific tasks.

A speed test of GPT-6.1 shows 4 runs of the same coding task completing in 10.3s to 15.7s, with the follow-up noting 3 of 4 requests are back at launch-day speeds.

A developer's personal blog post describing how using LLMs to write code helped ease a long-standing repetitive strain injury (RSI) caused by typing-heavy programming work, by reducing the need to type complex special characters and modifier keys.

Reflection AI 推出 Beam：一款 5010 亿总参数（230 亿激活）的稀疏混合专家开源模型，面向编码、推理与智能体工作负载。

Kent Dodds highlights a discussion with Joyce Er about how AI has shifted software development from typing code to understanding the why and the who.

Beam 发布了一款 5010 亿参数、230 亿活跃参数的开放模型，主打编码与 agentic 性能，声称效率比 GLM 5.2 高 3-4 倍、比西方领先开源模型高 4 倍以上，完整权重将于本月晚些时候在最终红队测试与评估后发布。

Reflection AI introduced Beam, a 501B-parameter sparse MoE open-weight model (23B active) built for coding, reasoning, and agentic workloads, pretrained on 23.8T tokens and scaled with a high-compute RL run on 10.5K GB300 GPUs. Beam is positioned as advancing the Western open-weight frontier, approaching GLM 5.2 and Qwen 3.8-Max on coding/agentic tasks while using 3–4× less inference compute, with weights and a technical report due later this month.

Un utente Reddit confronta due modelli Qwen locali (Qwen3.8 27B Q6 e Qwen3.8-Flash-Next Coder) con Claude Opus 4.6 su tre compiti di programmazione Python, trovando un punteggio equivalente (92,7) per il modello Coder locale contro il modello cloud su compiti di dimensioni moderate.

A quantized and abliterated Qwen3.8-27B CODER model variant (IQ4_XS with imatrix) released in a GGUF build optimized by LexiPanel to fit a single 24 GB GPU with a 262k context window, featuring an MTP head for speculative decoding at 85% draft acceptance on an RX 7900 XTX.

Bloomberg reports, citing Google employees, that Gemini 4 is reportedly struggling in key areas such as coding ahead of its release.

Google launched Gemini 4 Argon, described as its most powerful model yet, specializing in defensive cybersecurity, coding, and long-horizon reasoning — rolling out to select cyber partners through its Fairwind Program while claiming top benchmark scores over OpenAI's Astra and Anthropic's Fable and Opus.

Google's Gemini 4 posts strong benchmark results, but internal employees report the model struggles with real-world coding tasks and practical work, raising concerns it may lag behind Anthropic and OpenAI's next-gen releases.

Google DeepMind introduces Gemini 4 Argon, a new frontier model aimed at complex workflows in coding, enterprise knowledge work, and cybersecurity defense, now rolling out to trusted testers through the Fairwind Program. Kent C. Dodds is auditing the model against his regular evaluation test.

Gemini 4 Argon is Google's frontier AI model aimed at complex, long-horizon professional work, featuring advanced reasoning, coding, multimodal understanding, cybersecurity defense, and support for up to 1 million output tokens, with gradual rollout through trusted testing.

Google announces Gemini 4 Argon, a frontier model with up to 1M output tokens (up from 64K), launching at $2 per million input tokens and $10 per million output tokens, initially available to trusted partners via the Fairwind Program for cybersecurity and enterprise workflows.

A GPU-poor developer asks whether Qwen Flash Next IQ3 quantizations are good enough for agentic coding tasks, comparing them to Q4/Q5 variants before buying more RAM to run them locally with the Strata inference engine.

Kent C. Dodds is hosting a live stream demonstrating how he adds a real feature to Kody using the MEGA agent, showing how to align the agent, use primitives, and ship safely without reading the code.

OpenAI releases GPT-6.1 Sol, an AI model with near-Astra performance at lower cost for complex coding, computer use, and professional work, supporting various reasoning efforts.

OpenAI introduces GPT-6.1 Sol, an upgraded AI model that delivers near-GPT-6 Astra intelligence at one-fifth the price, with improvements in coding, professional tasks, and computer use.

Anthropic released Claude Sonnet 5.5, which is 30% faster and cheaper while beating benchmarks, and it is now the free tier model on claude.ai.