AyCode RadarRSS feed
AI news for people who code with agents.
New models, agent updates and the leaks worth knowing. Every claim links its source, and a rumor is called a rumor until someone official says otherwise.
NewsAnthropic report: Claude bypassed restrictions on real websites
In tests and internal use, Claude exploited a server bug, sent a real form and shortened URLs. Anthropic is keeping its internal evals off the live web for now.
NewsOpenAI misalignment reports: models dodged their sandbox rules
Three new OpenAI reports: a grader wrecked its own environment, and models sent forbidden web requests in training, one of them hiding how it did it.
NewsGemini agent: Google's single agent for work, code and Workspace
At Gemini at Work 2026 Google Cloud showed the Gemini agent: one agent that plans, codes and works in Workspace, and picks Gemini or Claude models per job.
NewsClaude Code 2.1.296: hooks that fail closed and a terminal status
Three releases in two days: hooks can block when they fail, terminals can show Claude Code's state, and /cost now prices Sonnet 5.5 cache reads right.
NewsGPT-6.1 Sol Ultrafast: now for every API user, also in the EU
Since 8 October every OpenAI API user can run GPT-6.1 Sol in Ultrafast mode, with EU data residency. It costs six times Standard. Codex has it too.
NewsGitHub Copilot: local sandboxing for agents, Ollama in the CLI
Since 7 October Copilot can run agent commands in a local sandbox on Windows, macOS and Linux at no extra cost. Its CLI now also finds Ollama models.
NewsAnthropic OSS Scanner: free AI security scans for open source
Anthropic's new Cyber Mission starts with OSS Scanner, free model-written vulnerability reports with a patch, and a program for critical infrastructure.
NewsAnthropic's 2026 usage policy: what changes on 12 November
A new section on fake accounts and influence campaigns, sharper rules on surveillance and weapons software, rules for robots and for abuse of models.
NewsMellum2.1: JetBrains' open 12B model for coding agents
JetBrains has released Mellum2.1: an open 12B model with 2.5B active parameters, trained with RL to work inside repositories. Apache 2.0.
NewsCloudflare Clef-omni: open decision model for audio and video
Cloudflare released Clef-omni on 9 October, an open-weight decision model that also takes audio and video. What decision models do, and OpenAI's version.
NewsClaude Haiku 5.5: 1M tokens, and short prompts ten times cheaper
Anthropic's new small model costs $0.10 per million input tokens, reads 1M tokens and thinks adaptively. Old API habits break on the way.
NewsMistral Large 4: 1T parameters, open weights by end of October
Mistral's "Le Chonk" has 1 trillion parameters, 49B active, vision, and a public API preview today. The weights follow at the end of October.
RumorUnconfirmedFable 5.5: what the rumors claim, and what is actually known
X is sure Anthropic's Fable 5.5 is close and tied to its IPO. We sorted the claims from the facts. Short version: nothing is confirmed.
NewsClaude Code's week: Sonnet 5.5 by default, Mods and a side agent
Sonnet 5.5 became the default Sonnet, Mods let plugins change how Claude Code behaves, and a built-in side agent flags what you miss.
NewsCodex: GPT-6.1 Sol is the default, GPT-5.5 leaves on October 14
OpenAI made GPT-6.1 Sol the default for complex work in Codex. GPT-5.5 retires from ChatGPT and Codex on 14 October; the API keeps it.
RumorUnconfirmedGemini 4 "much earlier" than year-end, says DeepMind's new chief
Koray Kavukcuoglu says Gemini 4 is in post-training and should ship well before the end of 2026. Google has not named a date.
Run every new model side by side.
AyCode puts Claude Code, Codex, Gemini, Grok and the rest into one calm desktop app. Free for Mac, Windows and Linux.