
Sakana AI's Fugu-Cyber Matches GPT-5.5 on the Hardest Security Benchmarks
Sakana AI's Fugu-Cyber matches GPT-5.5-Cyber on security benchmarks by orchestrating specialist agents -- but argues raw scores are only half the story
20 stories

Sakana AI's Fugu-Cyber matches GPT-5.5-Cyber on security benchmarks by orchestrating specialist agents -- but argues raw scores are only half the story

OpenAI open-sources Symphony, a spec for running unlimited Codex agents from your issue tracker — no babysitting required

Prefect ships FastMCP 3.0, a ground-up architectural redesign of the most widely used MCP framework, now powering 70% of all MCP servers

Cursor's new agent swarm rebuilt SQLite from its 835-page manual in Rust, passing 100% of tests at up to 8x lower cost than before.
Thinking Machines’ 975B open-weights MoE activates 41B parameters, ships Apache 2.0, and dials thinking effort without closed-API lock-in.

Unsloth now officially supports AMD GPUs with 2x faster training, 70% less VRAM, and a one-line installer across Windows, Linux, and WSL

Alibaba's Qwen-Audio-3.0-TTS tops the global TTS leaderboard with 16-language support, inline emotion tags, and noise-robust voice cloning

OpenBMB open-sources StaffDeck, a platform that turns business SOPs and knowledge into persistent, auditable AI agents with employee IDs and work records.

A new LoRA optimizer replaces Adam with three matrix-aware ingredients, hitting the same loss in 1.2 to 1.7 times fewer steps.

CloakBrowser patches Chromium's C++ source directly, passing 30+ bot detection tests including Cloudflare Turnstile and reCAPTCHA v3 with a 0.9 human score

A Go-based tool wraps AES-SIV encryption with a local LLM that turns your secret messages into innocent-looking chat text.

Epoch AI finds that style-mimicking AI fools leading detectors up to 29% of the time, with scientific writing as the biggest blind spot

Kimi K3 debuts at #3 on DeepSWE, matching Claude Fable and GPT-5.6 Sol as the first open-weights model to reach frontier-level coding performance at 2.8 trillion parameters.
AlphaSignal Signaldesk run: 13 tasks, 7 models, 488 attempts

Thinking Machines' Inkling becomes the top open-weight model on ARC-AGI-1 and ARC-AGI-2, scoring 79.5% and 36.5% respectively at under $1 per task.
A new paper accepted at ALIFE 2026 shows biologically constrained neural networks can learn competitively — without backpropagation's weight transport trick — across image classification and reinforcement learning.

LangChain's open-source deep research agent runs on any LLM and search tool, hitting #6 on the Deep Research Bench leaderboard with a fully configurable, MIT-licensed architecture.

Microsoft's open-source Data Formulator 0.7 brings agentic data exploration to enterprise sources — no SQL or coding required

Databricks signs a term sheet for a new round led by Coatue at a $188B valuation, up 40% from its last raise just months ago, to fund its multi-AI governance and agent infrastructure push.

LM Studio launches Bionic, an agentic AI for open models with local voice transcription, document editing, coding, and privacy-first cloud inference via Zero Data Retention.