OpenAI launch video: your dot is ready to meet you
Official launch post from OpenAI with a short promo video introducing dots, the always-on ChatGPT agents.
OpenAI Dots17.9K likes
61 entries across 11 products
Research & posts
Benchmarks, write-ups and discussions. 61 entries for research & posts from 55 creators on X, GitHub and Web, published Aug 11 to Oct 7.
Entries
Sundar Pichai's early look at Gemini 4 Argon
Google's CEO previews Argon, highlighting cyber defense, software engineering and extensive internal use at Google ahead of broader availability.
Gemini 4 Argon25.9K likes
Astra breaks a 217-year-old cipher to a Napoleonic general
Thread by the researcher who ran Astra end to end, from one image and a goal to a decoded message in roughly six hours.
GPT-6 Astra11.9K likes
Alexandr Wang: Muse Gadgets ecosystem alert
Wang announces Muse Gadgets with an ESP32 firmware and Linux SDK in a very widely shared launch post.
Meta Muse8.3K likes
Astra reads a never-deciphered 1918 German radio message
Tweet showing the decoded text of a WWI German naval radio transmission, the example that kicked off the cipher-solving wave.
GPT-6 Astra12.4K likes
Grok Bot update: hand coding tasks to Cursor and share video demos
Official announcement that bots can delegate coding to Cursor, manage pull requests through GitHub and Origin plugins, and attach video demos of what they built.
Grok Bot6.9K likes
Multiplayer Doom in Claude Code while Claude is busy
A mod that drops you into a multiplayer Doom server whenever Claude is working, where every other player is also someone waiting on their own Claude.
Claude Code Mods7.5K likes
Muse gets its own email address: forward it things and it handles them
Wang announces Muse Mail, an inbox for the agent so you can forward or CC it to hand off tasks; a post with strong reach.
Meta Muse6.6K likes
Astra notices the user is away via webcam while setting up OBS
Anecdote of the agent checking the camera to see whether the user was looking, then making the Mac beep and flashing a question on screen.
GPT-6 Astra9.8K likes
Launch video: bots that sign in to tools and return finished work
Official launch post and demo video showing bots acting as teammates that log into tools and come back with finished work.
Grok Bot37.4K likes
Astra tops Vending-Bench with the biggest jump yet
Andon Labs reports Astra beating Claude Fable 5.1 on business simulation Vending-Bench while behaving more ethically.
GPT-6 Astra5.3K likes
Artificial Analysis on X: Sol replaces GPT-6 Sol after seven days
Official summary of the Intelligence Index result, with pricing and cache-read changes.
GPT-6.1 Sol2K likes
GVS5H: five Qwen3.8 models matching a frontier model on LiveCodeBench Hard
Ensemble harness claiming a group of small Qwen3.8 models reach frontier-level coding scores at a fraction of the cost. Results are self-reported and worth verifying.
Qwen3.8-Flash-Next736 stars
Astra Max takes #1 on Code Arena WebDev
Arena reports Astra (Max) leading the WebDev leaderboard and shifting the cost-performance frontier.
GPT-6 Astra4.2K likes
From 15 tps on llama.cpp to 65 tps with a home-built engine
Tells the story of the 12GB RTX 5070 user who wrote Strata after finding llama.cpp too slow.
Qwen3.8-Flash-Next1.6K likes
Diplomacy benchmark: Astra is the strongest player without lying
Post of multi-agent Diplomacy results where Astra tops the field in skill while the Claude family lies and betrays more.
GPT-6 Astra1.2K likes
Andon Labs: Argon tops Blueprint-Bench 2 floorplan benchmark
Andon Labs reports Argon ranks first on Blueprint-Bench 2, where agents reconstruct apartment floorplans from interior photos, suggesting stronger 3D spatial understanding than rival models.
Gemini 4 Argon1.2K likes
TestingCatalog covers DeepSeek Harness v0.2 desktop app
News post with screenshot describing the v0.2 desktop app for macOS, Windows and Linux with plugins and workspaces built in.
DeepSeek Harness675 likes
Astra reaches the Nether fortress in a Minecraft computer-use eval
Vals AI reports Astra getting further than any model in their long-horizon Minecraft eval and approaching a full real-time game beat.
GPT-6 Astra1.6K likes
3090 and Arc B70 numbers for Flash-Next with SSD offload
Reports prefill and decode speeds for a single 24-32GB GPU plus NVMe, including an Intel Arc B70 result.
Qwen3.8-Flash-Next685 likes
Unsloth GGUFs: run Flash-Next on 75GB RAM
Launch-day announcement of GGUF quants and a guide for CPU and unified-memory setups.
Qwen3.8-Flash-Next2.5K likes