Skip to content

Search stories, models, companies…Subscribe

The river

Latest

Every story as it’s published, newest first.

Monday, Oct 12

1 story

Developers · ExplainerClaude Code's "Compacted while idle" message, explained: why it happens and how to turn it offSince about version 2.1.286, Claude Code can summarize a large conversation on its own shortly before its one-hour prompt cache expires. Why Anthropic does it, when it fires, what it costs you, and the one-line setting that stops it.15:29·9 min read

Friday, Oct 9

29 stories

Agents · NewsThe Week in AI #2: Sonnet 5.5, OpenAI DevDay and a gated Gemini 4September 28 – October 4, 2026: two $2/$10 workhorse models, OpenAI's always-on dots, Google's cyber-first Gemini 4 Argon, and Copilot learns to use your desktop.11:14·6 min read

Agents · NewsThe Week in AI #1: Opus 5.5, GPT-6 Sol and Grok 4.7 land within 48 hoursSeptember 21–27, 2026: three new coding-focused frontier models, Cursor's deploy-watching bots, and a sandbox for GitHub Copilot's desktop app.11:14·4 min read

Open source · NewsDeepSeek-V4.1-Flash: an MIT-licensed open-weights MoE that DeepSeek says beats its own flagshipReleased September 10, DeepSeek's new Flash model activates 8B–16B parameters per token, reads 1M tokens and images, and now answers V4-Pro API calls. What's confirmed and what's vendor-reported.11:14·7 min read

Industry · NewsOpenAI reportedly seeks $30 billion at a $1.4 trillion valuation as it deepens Atlassian dealThe funding talks are unconfirmed and OpenAI's IPO has slipped to 2027. The Atlassian expansion, announced October 6, is the part OpenAI has put in writing.11:13·4 min read

Policy · NewsChatGPT starts watermarking text in the EU as the AI Act's marking deadline approachesOpenAI's October 5 rollout is the most visible sign yet of the AI Act's content-marking rule. Systems already on the market have until December 2 to comply.11:13·5 min read

Hardware · NewsNvidia's Vera Rubin reaches customer production as AMD buys an AI labCoreWeave says Cognition is the first customer running production work on Vera Rubin NVL72. Two days earlier, AMD agreed to buy World Labs for about $8.2 billion in stock.11:13·4 min read

Products · NewsGemini's new skills are replacing Gems. Here's what changes and whenGoogle's reusable instructions can be built from your chats, called with a slash and combined. Gems for personal accounts go away in November.11:13·5 min read

Developers · NewsOpenAI's Agents API puts the Codex harness behind one endpoint: what developers get, and what they give upIn public beta since September 10, with computer use added September 29. How sessions, sandboxes and approvals work, what it costs, and the catches.11:12·5 min read

Developers · NewsCursor's latest release puts your local agents on your iPhoneRemote control for local agents arrived on October 6, capping a month of agent-management features in Cursor. What's new, what it costs, and what's unclear.11:12·6 min read

Developers · NewsGitHub Copilot can now drive your desktop apps: what computer use does and how to turn it onComputer use reached public preview on October 1 in Copilot CLI and the Copilot app on macOS and Windows. What it does, what it costs, and the risks GitHub flags.11:12·6 min read

Models · NewsGoogle's Gemini 4 Argon is here, but only cyber defenders can use it so farGoogle announced its new frontier model on September 30 with a $2/$10 introductory API price. A week later, access is still limited to vetted security partners.11:12·6 min read

Models · NewsGPT-6 reaches every ChatGPT user: what OpenAI shipped, what it costs, and where each model runsFour GPT-6 models in five weeks, split across ChatGPT's Chat and Work tabs, Codex and the API. Here's the verified map, with prices.11:12·6 min read

Research · AnalysisOpenAI's Navier–Stokes claim, explained for builders: the verifier is the storyOpenAI says 10,000 AI agents found a blowup in the Navier–Stokes equations and proved it in Lean. What is checked, what isn't, and what it means for agent builders.11:12·6 min read

Models · AnalysisWhat the Claude Opus 5.5 system card actually saysAnthropic's system card for Opus 5.5 runs past 220 pages. What it says on bio and cyber risk, sandbox escapes, prompt injection and alignment, and what it leaves out.11:12·7 min read

Industry · AnalysisToken pricing isn't the story: what the official price pages actually tell youAnthropic and OpenAI now charge identical sticker prices at two tiers. Your bill is decided by tokenizers, thinking tokens, caching, long-context surcharges and tool fees.11:12·8 min read

Open source · AnalysisOpen-weight licenses compared: Llama vs Apache, MIT and the new revenue-gated licensesWe read the license texts. What Llama, Apache 2.0, MIT, Gemma's terms and the newest Qwen and Mistral licenses actually let you do, and why "open weights" isn't "open source".11:12·8 min read

Hardware · AnalysisThe gigawatt datacenter race: what's announced, what's running, and what's holding it backAI labs and clouds have announced tens of gigawatts of compute. Primary sources show far less of it running. A sourced table and an honest look at the limits.11:12·8 min read

Agents · AnalysisThe case against unattended agents: what went wrong, and how sandboxing actually worksFrom a deleted production database to poisoned npm releases, documented agent failures share a pattern. Here's what the sandboxes in Claude Code, Codex and Copilot actually enforce.11:12·9 min read

Models · AnalysisDo AI labs really ship 'mini' models first? What the release record since 2024 showsAcross 32 small-vs-flagship pairings from six labs, the small model came first only five times. Here's what really drives the timing, and the price war at the bottom.11:12·9 min read

Models · BenchmarksLong-context recall: what independent benchmarks say "1 million tokens" really buys youEvery frontier API now accepts about a million tokens. Independent evaluations from RULER to ATLAS show how much of that window models actually use.11:12·9 min read

Agents · BenchmarksSWE-bench, Terminal-Bench and METR: what coding-agent evaluations measure, and what their results don't proveOpenAI dropped SWE-bench Verified, Terminal-Bench is on version 4.0, and METR can't yet pin down AI's effect on developer speed. What published evaluations show, and their limits.11:12·8 min read

Industry · CompareFrontier APIs vs open models: what they really costHosted open models are cheap, but the frontier labs' small models now list even lower. Self-hosting pays off only at very high, steady volume. Verified prices and the arithmetic.11:12·Checked Oct 8

Products · CompareChatGPT vs Gemini for research: deep research, long documents, prices and privacy comparedBoth now write cited research reports. The real differences are how much each plan can read at once, how they handle your PDFs, and what happens to your data.11:12·Checked Oct 9

Products · CompareChatGPT vs Claude for coding and writing: plans, models and limits comparedBoth start at $20 and go to $200 or more. We compare the models each plan gets, Codex vs Claude Code, context, limits and what leaderboards show.11:12·Checked Oct 9

Agents · CompareClaude Code vs Codex vs Cursor: features, prices and limits compared (October 2026)All three now run in the terminal, the IDE and the cloud. The real differences are how they meter usage, how they sandbox commands, and which models you get. Verified against official pages.11:12·Checked Oct 9

Hardware · ExplainerWhat an NPU actually does, and why the TOPS number on your laptop matters less than you thinkCopilot+ PCs need a 40-TOPS 'AI engine'. Here's what that chip accelerates, which apps really use it, and why tools like Ollama mostly ignore it.11:11·7 min read

Products · ExplainerHow Gemini handles a 900-page PDF: the limits, the costs and what long-context benchmarks showOn paper, a 900-page document fits comfortably in Gemini's 1M-token context. Whether it works depends on file size, your plan and what you ask.11:11·6 min read

Products · ExplainerThe Claude app, explained: plans, models, Cowork and how usage limits workClaude now covers chat, the Cowork agent, a Chrome extension and Claude Code under one subscription. Here's what each plan gets in October 2026.11:11·7 min read

Products · ExplainerChatGPT agent mode, explained: what it was, why it's gone, and what ChatGPT Work does insteadOpenAI's help center says ChatGPT agent is no longer available. ChatGPT Work, a cloud browser and dots now do its job. Here's how they work.11:11·7 min read