AI Coding Live News

X Discussion: levelsio Used Claude Code to Decompile the Battlefield 1942 Demo and Put It in the Browser With Multiplayer, Citing About 1M Output Tokens

Indie developer Pieter Levels (@levelsio) said on X on November 2 that he used Claude Code to decompile the engine of the free Battlefield 1942 demo, port it to C, compile it to WebAssembly and host it at bf1942.pieter.com, where it runs in the browser with online co-op on the Tobruk map. He said the job took about 1 million output tokens, roughly $100 at API prices, while he pays for a $100-a-month Claude subscription. His two main posts drew about 2,200 likes and 340,000 views and reached X trending. These are the author's own claims and community discussion, not independently verified.

X Discussion: levelsio Used Claude Code to Decompile the Battlefield 1942 Demo and Put It in the Browser With Multiplayer, Citing About 1M Output Tokens

Key Takeaways

  • Author's account: Claude Code decompiled the Battlefield 1942 demo engine, which was ported to C and compiled to WebAssembly to run in the browser.
  • The live build at bf1942.pieter.com opens directly, is labeled non-commercial and unofficial, and offers online co-op on Tobruk with no install.
  • Cost: about 1 million output tokens by the author's count, roughly $100 at API prices; he actually pays for a $100-a-month Claude subscription.
  • He said it was much harder than his earlier Quake and DOOM ports, whose engine source code id Software released long ago.
  • Reach: the first post had about 1,100 likes and 230,000 views, the launch post about 1,100 likes and 107,000 views; testers are already reporting bugs.
Read full story
Linked Models & Agents:

Vercel Labs launches agent-desktop: computer use for any coding agent via OS accessibility trees, not pixels, on macOS, Windows and Linux

On October 31, Chris Tate of Vercel Labs, creator of agent-browser, announced agent-desktop on X as "agent-browser, for your whole computer". It is an Apache-2.0 Rust CLI and C library that reads a desktop app's OS accessibility tree and returns every actionable element as a JSON ref; the agent clicks or types on a ref and snapshots again to verify, with no pixel matching, headless by default. Version 1.0.0 shipped on October 27 and added Linux support; the current npm release is 1.0.2.

Vercel Labs launches agent-desktop: computer use for any coding agent via OS accessibility trees, not pixels, on macOS, Windows and Linux

Key Takeaways

  • Install: npm install -g agent-desktop, one package with native x64/ARM64 binaries; 1.0.0 (Oct 27) added Linux (glibc 2.35+, via AT-SPI), and the latest npm release is 1.0.2 (Oct 30)
  • Loop: snapshot prints a JSON tree with refs, click @<snapshot_id>:eN acts, then snapshot again to verify; a changed UI returns STALE_REF, and errors say whether a retry is safe
  • Scope: the README lists 60 command names, 56 of them operational, spanning windows, keyboard and mouse, notifications, clipboard, waits and session traces
  • Headless by default: ref actions go through accessibility APIs without moving the pointer or stealing focus, with --headed for physical input; Chris Tate says multiple agents can use the computer in parallel, each with its own cursor
  • Benchmarks: no official success-rate or speed numbers; it is a tool with no model inside, and web pages stay with agent-browser
Read full story

Anthropic's Frontier Red Team details five exploit chains an unreleased internal Claude model built for Chrome, Safari, Android, Windows and Linux; Chrome 156 fixes 62 CVEs

Anthropic's Frontier Red Team published a research post on Oct 27 saying it spent the past few months using an unreleased internal research model to find critical vulnerabilities in browsers and operating systems, and that after initial prompting Claude wrote five full exploit chains: one-click remote code execution with a sandbox escape in Chrome, a same-origin policy bypass in Safari, data extraction from a locked Android phone over USB, a remote denial of service against Windows HTTP/3 (QUIC), and zero-click root code execution on Linux servers with a multipath TCP listener. Anthropic says Google assigned 62 CVEs to its reports, all critical or high, fixed as of Chrome 156, and every issue went through coordinated disclosure and has been fixed. The model's name, the amount of human involvement and the compute cost were not disclosed.

Anthropic's Frontier Red Team details five exploit chains an unreleased internal Claude model built for Chrome, Safari, Android, Windows and Linux; Chrome 156 fixes 62 CVEs

Key Takeaways

  • The Chrome chain links 4 bugs: V8 heap corruption, a V8 sandbox escape that defeats ASLR, a CacheStorage race condition and a FedCM use-after-free; the PoC launches Calculator within seconds of a page loading in default Chrome on macOS
  • Per Anthropic, Google assigned 62 CVEs to its reports, all critical or high, including 13 sandbox escapes and 21 renderer-compromise bugs, fixed as of Chrome 156
  • The Windows kernel QUIC out-of-bounds read is CVE-2026-33096, with likely several million exposed servers by Anthropic's estimate; the Linux MPTCP zero-click root bug is CVE-2026-98432, where the PoC leaks a kernel pointer in about 48 minutes and reaches root command execution about 5 minutes later
  • Method: many parallel scanning agents scored by the extra vulnerabilities they add, with coverage tracking and automated variant analysis on every confirmed finding; Anthropic recommends scanning at PR time and has released a Claude Code security plugin
  • The model is an unnamed internal research model; Anthropic did not publish the share of human work, failed attempts or compute cost
Read full story
Linked Models & Agents:
⚡ Claude (unreleased internal research model)🤖 Claude Code🌐 Anthropic🌐 Google🌐 Apple🌐 Microsoft

Cursor Design Mode adds Annotate: point, draw and edit text on your running app to send visual prompts to agents

On Oct 27 Cursor announced Annotate for Design Mode: in the Agents Window browser you can click elements, box an area, draw markup, edit text in place or tweak CSS, and each note is sent to the agent with element details and a screenshot so it can edit the code. The post says you can also review designs at different screen sizes. It needs Cursor 3.24.18 or later, and the shortcut is Cmd+Shift+D.

Cursor Design Mode adds Annotate: point, draw and edit text on your running app to send visual prompts to agents

Key Takeaways

  • Requires Cursor 3.24.18 or later; click Annotate in the Agents Window browser toolbar or press Cmd+Shift+D
  • Each note carries the element tag, id, classes, selector path, first 200 visible characters, size and position, six computed styles and a cropped screenshot (official docs)
  • React, Vue and Svelte apps also send the component name and component stack so the agent can find the source
  • Batch edits switch: off sends each note on Return, on collects numbered pins and sends them as one message
  • Cursor recommends a fast model that is strong at interface work, such as Composer 2.5
Read full story
ADSponsored

Cursor launches Computer Use: agents can control your computer and drive desktop apps like the iOS Simulator to test what they build

On 2026-10-23 (15:59 UTC) the official Cursor account @cursor_ai announced "Cursor can now control your computer": agents can drive desktop apps such as the iOS Simulator to test the features they build. A follow-up post says Computer Use lets agents click and type across any app on your computer, not just the browser. Cursor turns it on by itself when a task needs it, and you can also request it by @-mentioning Computer Use in your prompt; updating Cursor is enough to try it. A community changelog in September mentioned Computer Use on Mac in version 3.20.21, and this is the official launch. Supported platforms, the permission model and benchmarks were not published.

Cursor launches Computer Use: agents can control your computer and drive desktop apps like the iOS Simulator to test what they build

Key Takeaways

  • Cursor officially says its agents can now control your computer and drive desktop apps like the iOS Simulator to test the features they build
  • Computer Use lets agents click and type across any app, not just the browser
  • Cursor uses it automatically when a task needs it, and you can request it by @-mentioning Computer Use in a prompt
  • Update Cursor to try it; supported platforms, the permission model and benchmarks were not published
Read full story

Gradium open-sources Phonon on-device TTS: a 31M-parameter, 64 MB model that starts speaking in 7 ms on a laptop CPU, with the English version under Apache 2.0

On Oct 20, French voice AI company Gradium open-sourced Phonon, a family of on-device text-to-speech models: the English Phonon 31m (64 MB q8 weights, Apache 2.0) and Phonon 40m (single-language English, French, German, Spanish and Portuguese variants plus a five-language model, CC BY-NC non-commercial). Built on Kyutai's Pocket TTS and the Mimi codec, it ships its own runtime for CPU, Metal, CUDA, Vulkan, WebGPU, Core ML and Snapdragon, running offline on phones, browsers and a Raspberry Pi. Gradium reports 7 ms to first audio on a MacBook Air M5 CPU versus 517 ms for Kokoro.

Gradium open-sources Phonon on-device TTS: a 31M-parameter, 64 MB model that starts speaking in 7 ms on a laptop CPU, with the English version under Apache 2.0

Key Takeaways

  • Check the license: only the English Phonon 31m is Apache 2.0 for commercial use; every Phonon 40m variant, including multilingual, is CC BY-NC 4.0 and needs a deal with Gradium for commercial use
  • Size (official): 64 MB q8 weights for 31m and 74.6 MB for 40m; 37 built-in English voices, 102 in the multilingual model, plus voices cloned or designed with Gradium
  • Speed (vendor-measured): 6.9 ms to first audio and 29.9x real time on a MacBook Air M5 CPU with 4 threads, versus 517 ms for Kokoro (which does not stream); 38 ms and 3.7x real time on a Raspberry Pi 5; 7.5 ms on Snapdragon 8 Elite Gen 5
  • Accuracy (vendor-built hard-sentence set, 500 per language, public on HF): English WER of 1.70% for 31m and 2.22% for 40m versus 3.50% for Kokoro and 8.71% for NeuTTS Nano, with text normalization for numbers, dates and version strings
  • Get started with uvx ptts or pip install ptts, the phonon-tts npm package for browsers, or the gradium-ai/xn-ptts runtime on GitHub
Read full story
Linked Models & Agents:
⚡ Phonon 31m⚡ Phonon 40m⚡ Phonon 40m multilingual⚡ Pocket TTS⚡ Kokoro🤖 Pipecat🌐 Gradium🌐 Kyutai🌐 Arm🌐 Qualcomm🌐 Hugging Face

Viggle releases JST-2, a physics-native world foundation model: 3D scenes in about one second and animation-ready characters from a single image

Viggle released JST-2 on October 19. Instead of learning in pixel or video space, it represents the world as persistent "atoms" in 3D space (Physics-Native Tokens) and uses one architecture to generate scenes, characters and motion. Viggle says a 3D scene takes about one second on an NVIDIA B200 at under $0.01 of inference, with up to 64 input views kept as spatial context, and characters animate without a separate rigging step. It is free to try in the PINOC browser app and is also available via API and MCP.

Viggle releases JST-2, a physics-native world foundation model: 3D scenes in about one second and animation-ready characters from a single image

Key Takeaways

  • Scene generation: about 1 second per scene (NVIDIA B200, excluding network latency) at under $0.01 inference cost; Viggle cites about 5–6 minutes for Marble (vendor figures).
  • Spatial context: up to 64 input views, expandable one view at a time in real time, output as 3D Gaussian splats rendered locally in the browser.
  • Character blind tests (run by Viggle): Average Score 58.4% vs Meshy 7.1 and 73.2% vs Tripo H3.1; face quality 63.0% and 82.0% (50% means parity).
  • Motion blind tests (run by Viggle): text-to-motion 74.8% vs Cartwheel Swing and 80.0% vs HY-Motion-1.0; video-to-motion 89.1% vs DeepMotion Animate 3D.
  • Access: free in the PINOC browser app, plus API and MCP for agents such as Codex and Claude Code; JST-3 aims to replace rule-based collisions and object interactions with learned physics.
Read full story
Linked Models & Agents:

Skywork open-sources Matrix-Game 4.0: game engine owns state, a 5B causal video model renders playable games in real time on one GPU

Skywork AI (Kunlun Tech) open-sourced Matrix-Game 4.0 on 2026-10-15. An executable game engine owns world state, physics and rules and emits a cheap low-poly RGB proxy; a 5B causal video model built on Wan2.2-Fun-5B-Control re-renders it into finished frames with 1–2 denoising steps. Inference code, a WebRTC browser demo and game source (Apache-2.0) are on GitHub, with Castle, Ancient Egypt and City Racing weights on Hugging Face. Skywork claims 480p/60 FPS and 112 ms end-to-end on an RTX 5090; the repo's own A100 measurements are ~38–39 FPS. Training code and data are not released.

Key Takeaways

  • Scale: each scene checkpoint has 5,001,016,512 BF16 params on an alibaba-pai/Wan2.2-Fun-5B-Control base, output 480×864 (HF card)
  • Vendor claim (RTX 5090): 480p at 60 FPS, 112 ms end-to-end, up to 83.4 FPS generation; repo-measured on one A100-80GB: ~38–39 FPS for Castle/Egypt fixed prompts, ~28.9 FPS for Racing dynamic prompts (v4 README)
  • Presets: realtime = TAEHV codec + 1-step denoise + CUDA graphs; quality = Wan VAE + 2 steps; 21-latent attention cache incl. sink, 4 frames per chunk
  • License: code and scene weights Apache-2.0; no training code or data; weights need the Matrix-Game 4.0 runtime (not drop-in Diffusers)
  • Design: a harness where the engine owns state and rules, a coding agent turns natural-language requests into new rules and prompts, and the video model only renders
Read full story
Linked Models & Agents:
⚡ matrix-game-4.0⚡ wan2.2-fun-5b-control🌐 skywork🌐 kunlun-tech🌐 huggingface

Web$hell Unifies Agentic Web Access: Bridging Search, Browser Interaction, and File Caching Cuts Deep Research Turns by 66%

As LLM agents are deployed on long-horizon, end-to-end open web tasks such as deep research, multi-vendor comparison, and cross-site navigation, existing web access paradigms suffer severe fragmentation: search-augmented agents discard interactive controls, browser-based agents incur prohibitive context token consumption, and filesystem-based tools (Fetch-then-Explore) lose crucial interaction contexts when selectively grepping text. Researchers from Yonsei University introduce Web$hell (arXiv:2610.13416), a unified interface standardizing web access. Anchored by WebContract (establishing consistent schemas for addresses, values, and associations), WebCompiler (deterministically transforming HTML semantics into operable information and action objects without LLM inference overhead), and WebCache (maintaining persistent, stateful page revision trees), Web$hell unifies passive discovery with active interaction. Evaluated across six benchmarks spanning Web Search (WideSearch, SealQA), Web Navigation (WebArena-Verified, WebWalkerQA), and Deep Research (DeepResearchBench, DeepWeb-Bench), Web$hell achieves state-of-the-art performance across all tasks. Crucially, it slashes Deep Research turns from 52.4 and 30.9 down to 17.6 (-43% to -66%), maintaining flat cost curves even when subjected to 400 injected distractor pages.

Key Takeaways

  • Web$hell unifies web search, interactive navigation, and file caching under a single interface for LLM agents
  • Employs WebContract, deterministic WebCompiler, and stateful WebCache to achieve SOTA across 6 web benchmarks
  • Cuts Deep Research interaction turns from 52.4/30.9 to 17.6 (-66%), demonstrating flat costs under 400 distractor pages
Read full story

ChatGPT Browser Annotation API goes extensible: custom selection and controls, paired with WebMCP site tools for agent actions

On 2026-10-12 @OpenAIDevs highlighted ChatGPT desktop browser Annotations extensibility (document.oai.annotation): sites can customize selection targets, metadata and preview controls, and pair them with WebMCP site tools so agents receive the user's selected page context and can act through the site's own capabilities. Docs target DevDay 2026+ desktop; Enterprise/Edu need admin policy.

ChatGPT Browser Annotation API goes extensible: custom selection and controls, paired with WebMCP site tools for agent actions

Key Takeaways

  • Docs: Annotations Extensibility and Site tools / WebMCP; API at document.oai.annotation
  • Surfaces: selection containers, text drag-select, metadata (≤6 keys / ≤2048B), request(), registerControls() (≤12), canvas registerSurface()
  • Split: Annotations bring selection + feedback into chat; WebMCP registers page actions; requires GPT-5.6 Sol or GPT-6 Sol (Luna has WebMCP off)
  • Limits: secure top-level pages only; Enterprise/Edu policy; no iframe tools / declarative HTML WebMCP; Edu has no Site tools
  • Verified post: OpenAIDevs 2109716341490106869
Read full story
ADSponsored

REMORY: HKUST and Alibaba Introduce Residual Memory for Agent Context Compaction, Matching Full Context with Only 5.2% Token Overhead

As autonomous agents execute complex multi-step workflows across terminal consoles and web browsers, context window expansion and escalating token costs impose severe operational bottlenecks. Conventional compaction methods rely on text summarization or token pruning, both of which discard crucial state variables and induce semantic drift over extended sessions. Researchers from HKUST and Alibaba present REMORY (arXiv:2610.11287), introducing a residual memory paradigm designed for sequence-level context compaction. REMORY combines lossy natural language summaries with bounded sets of learned soft memory tokens, effectively forming a residual connection in the sequence dimension. While text summaries convey high-level narrative continuity, the soft memory tokens reconstruct lost fine-grained execution artifacts, API arguments, and intermediate environment states. On challenging benchmarks including BrowseComp and Terminal-Bench 2.1, REMORY retains 97% of full-context task performance while occupying only 5.2% of the original input token footprint, slashing redundant tool retries and loops by more than 40%.

Key Takeaways

  • HKUST and Alibaba propose REMORY, establishing sequence-dimension residual memory connections to resolve agent context bloat
  • Matches 97.2% of full-context capability using only 5.2% of token positions, slashing repetitive tool errors by 41.5% on Terminal-Bench
  • Non-invasive plug-and-play architecture reduces TTFT latency by 73.8%, setting a new benchmark for long-horizon agent efficiency
Read full story
Linked Models & Agents:

Pine Computer launches a cloud computer built for AI agents: event-driven perception instead of screenshot polling; GPT-5.6 Luna hits 78.3% SaaS-Bench v1.1 checkpoint score vs 74.3% for Opus 5 + Claude Code at ~$1.02 model cost per task

On Oct 9 Pine AI released Pine Computer in private beta: cloud virtual computers plus a harness and runtime built for AI, driven by an API across browser, files and shell. Instead of repeated screenshots, the OS, browser and apps push change notifications to the model ("epoll for AI perception"). On SaaS-Bench v1.1 (106 tasks, 23 apps) GPT-5.6 Luna on Pine scored 78.3% on checkpoints vs 74.3% for Opus 5 + Claude Code, though its fully-resolved rate (27.4%) trails the latter (31.1%).

Pine Computer launches a cloud computer built for AI agents: event-driven perception instead of screenshot polling; GPT-5.6 Luna hits 78.3% SaaS-Bench v1.1 checkpoint score vs 74.3% for Opus 5 + Claude Code at ~$1.02 model cost per task

Key Takeaways

  • SaaS-Bench v1.1 checkpoint score: Pine + GPT-5.6 Luna 78.3% vs Opus 5 + Claude Code 74.3% vs GPT-5.6 Sol + Codex 71.1%
  • Fully resolved rate trails: 27.4% vs 31.1% (Claude Code) / 29.2% (Codex)
  • Model token cost per task: ~$1.02 vs $26.50 (Opus 5 + Claude Code) / $20.50 (GPT-5.6 Sol + Codex), excluding infrastructure
  • New computer boots in seconds, heavy state restores in ~15 s, paused ones resume in <1 s
  • Launch post: ~99.8K impressions, 250+ likes, 120 reposts; private beta waitlist only
Read full story

Odyssey launches Odyssey-3 world model: new SOTA 66.1 on Physics-IQ Verified, free real-time browser preview, and robot-arm control

On Oct 8 Odyssey released Odyssey-3, a foundation world model built as an autoregressive diffusion transformer. Odyssey-3 Pro (720p) scores 66.1 on Physics-IQ Verified video-to-video, the highest reported score, and 54.7 image-to-video. A free browser research preview (Flash variant) offers first/third-person navigation and free camera; API access is by request, with no open weights or public pricing. Odyssey says tens of hours of demos are enough to control various robot arms.

Odyssey launches Odyssey-3 world model: new SOTA 66.1 on Physics-IQ Verified, free real-time browser preview, and robot-arm control

Key Takeaways

  • Physics-IQ Verified v2v: Odyssey-3 Pro 66.1 (best-of-8, highest reported) / 63.4 prompt-enhanced; 480p model 51.8 → 61.6 → 64.4 (launch post)
  • Image-to-video: Odyssey-3 Pro 54.7; vendor says 1st in 3 of 4 WorldMark categories (self-reported)
  • Architecture: autoregressive diffusion transformer predicting environment evolution in real time from past observations plus latest actions/events
  • Physical AI: an action decoder trained on tens of hours of demos controls robot arms, with recovery behaviors absent from training data (technical write-up)
  • Access: free browser research preview; API by request, no open weights or public pricing
Read full story
Linked Models & Agents:
⚡ Odyssey-3⚡ Odyssey-3 Pro⚡ Odyssey-3 Flash🌐 Odyssey

Cline launches Cloud Agents: assign tasks in the browser, cloud sandboxes code/test and open PRs, up to 10 in parallel

Cline moved its open coding agent into the browser: connect GitHub, pick a repo and model, and each task runs in an isolated cloud sandbox that commits to a cline/ branch and can open a PR—even with your laptop closed. Limits: 10 concurrent sessions and 50 new sessions/day. Providers: ClineFree, ClinePass ($9.99), and usage-billing. Cloud sessions do not yet load Rules, Skills, MCP, or Connectors.

Key Takeaways

  • Surface: no install; assign work from app.cline.bot Agents; each session gets its own sandbox and repo copy (docs)
  • Delivery: continuous commits on a dedicated cline/ branch, then optional PR; idle 15 minutes pauses the sandbox, resume keeps the workspace
  • Scale: up to 10 concurrent sessions and 50 new sessions/day; mobile browser for approvals/follow-ups
  • Models: ClineFree trials, ClinePass $9.99/mo (~2–5× open-model usage), usage-billing; switch mid-session
  • Security / gaps: GitHub token scoped to the single chosen repo; cloud does not load Rules/Skills/MCP/Connectors yet (local IDE/CLI/Desktop remain full-featured)
Read full story

Grok Bot adds a Shopify connector for orders, inventory and listings; Shopify announces @grok and @bot connectors

On 2026-10-08 @bot said Grok Bot now connects to Shopify to check orders, track inventory, and update listings; Shopify announced matching connectors for @grok (business Q&A) and @bot (agent teams). A structured connector avoids the browser bot checks that blocked earlier workflows. Scopes and write operations are not yet detailed.

Grok Bot adds a Shopify connector for orders, inventory and listings; Shopify announces @grok and @bot connectors

Key Takeaways

  • Official: @bot and @Shopify announced the same day (2026-10-08)
  • Capabilities: orders, inventory tracking, product listings
  • Shopify also launched a @grok connector (business Q&A) alongside @bot (agent teams)
  • Structured connector instead of browser login, avoiding earlier 'Verify Human' blocks
  • Not disclosed: scopes, refunds/discounts, quotas; not yet in the Grok Bot changelog at ingest
Read full story
Linked Models & Agents: