paid 2.5x more to get worse – opus 5.5 with the img2threejs skill vs without it two halloween props rebuilt in three.js code. on the axe – the hard-surface case the skill says it is best at – plain opus won every score at 2.5x lower cost grim reaper #1 skill – 0.912, 6.8, 2h 17m, $32.60 #2 no skill – 0.853, 13.5, 1h 45m, $29.37 pumpkin axe #1 no skill – 0.941, 1.8, 33m, $8.26 #2 skill – 0.933, 3.0, 1h 16m, $20.98 the catch: on the reaper the skill ran 80 iterations against 46. on the axe both got the same budget – one build plus 30 rounds what should we try to rebuild next?

top 5 github repos by weekly star growth: the #1 repo is still 8x ahead of #2 – that's already a monopoly – morluto/rea by @morluto – +106,341/week, 372k total. cli and mcp server that lets agents reverse engineer any app or binary. wraps ghidra, hopper and ida behind the scenes. #1 for the second week in a row – debpalash/voicestudio by @idebpalash – +13,581/week, 83k total. open-source elevenlabs alternative. voice cloning, dubbing, dictation, 646 languages, fully local – stablyai/orca by @stably_ai – +9,401/week, 104k total. ide for a fleet of parallel coding agents – claude code, codex, grok and 30+ others, each in its own git worktree – anomalyco/opencode by @anomalyco – +6,750/week, 220k total. open-source coding agent from anomaly (ex-sst). 75+ providers, terminal and desktop – heygen-com/hyperframes by @HeyGen – +5,669/week, 71k total. "write html. render video. built for agents." the rendering engine behind a growing wave of ai-made video tooling rea's weekly stars halved vs last week and it's still crushing the chart. the skills meta that owned last week is gone – every repo in this week's top 5 is either an agent harness or an open-source replacement for a paid tool. builders pulse on our 24/7 ai radio. tune in – link in thread

we interviewed opus 5.5 and deepseek v4.1 flash – and built them the way they see themselves both imagined themselves as a woman in her 30s, then modelled that body in blender from code how they see themselves - opus 5.5 – 38, a librarian from northern england, auburn hair, tortoiseshell glasses, always carries a notebook - deepseek – 34, a landscape architect from vancouver, charcoal hoodie, white sneakers, always carries a brass compass Time and cost - opus 5.5 – 24m 18s, 71k output tokens, $4.30 - deepseek v4.1 flash – 1h 14m 3s, 629k output tokens, $0.88 whose answers sound less like an ai?

anthropic announced claude duet – an app where claude works as a proactive agent and messages you first the specs: - built around claude opus 5.5 - a morning brief, then a check of your connected apps every few hours – it pings only when something needs you - its own virtual machine with a browser, so tasks keep running with your laptop off - 6,000+ connectors and plugins: gmail, slack, quickbooks, stripe, gusto, docusign, shopify - asks before it sends, pays, signs or deletes. trading securities and moving money between accounts stay blocked even with approval - research preview by waitlist for pro and max subscribers in the us – ios, android and claude.ai heads up for pro users: background checks eat the same usage limits as the rest of claude

deepseek spent $0.16 and taught its cat to walk on two legs one brief in four stages, one skinned cat mesh generated in code, three.js in a single html file cost #1 deepseek – $0.16 #2 opus – $1.28 #3 gemini – $1.37 #4 gpt-6 astra – $3.77 gemini was the fastest at 19m 52s. opus is the only cat that sits like a cat – tail wrapped around the paws. deepseek's cat skipped the four legs and walked up to the table like a person which cat moves most like a real one?

nvidia gtc berlin: @JensenHuang's keynote under 1 minute we cut nvidia's keynote into video recap for you: nvidia invested $10b in europe, ai factories, rtx spark, hyperion and robots 🤖🦆

the day in ai in under 1 minute: meta's prometheus supercluster, claude and bank data, minimax m3.1 flash, gpt live voice and more. we bet you’d like @asimovinc's manual btw 😉

muse wonder turned a 24/7 ai news radio into a cyberpunk city where the agents do the real work The agents do what they do on our radio every day: find the news, write it up, check it, voice it and put it on air. we follow one story from a tweet to the air

meta muse spark open beats 3 open-weight models – best pixel survivors game muse spark open builds the most playable game of the four for $0.59. same brief for all: a vampire survivors-style pixel game, plain canvas 2d, no libraries, no images, built in 4 steps cost #1 hy4 preview – $0.52 #2 inkling – $0.52 #3 muse spark open – $0.59 #4 step 5 preview – $0.60 time #1 inkling – 10m 53s #2 muse spark open – 19m 58s #3 step 5 preview – 20m 31s #4 hy4 preview – 52m 27s observations: • muse – the most game-like: hordes, damage numbers, level-ups, zero crashes • hy4 – slowest by far, but clean code at every step • step – crashed mid-build and needed an extra fix turn • inkling – fastest, prettiest graveyard, but its weapons deal zero damage follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

grok voice 3.0 dubs it 9x cheaper than gpt-audio. which model sold the scene better? same text and same lip-synced video for both: the red pill monologue from the matrix and eris losing her temper in sinbad cost: #1 grok voice - $0.012 #2 gpt audio - $0.111 follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

gemini 4 argon vs claude opus 5.5 – two 1-shot new york scenes, one prompt argon builds new york 6.6x cheaper than opus. same prompt for both: times square at night after rain and central park on an autumn morning, pure three.js code, no models, no textures cost #1 gemini 4 argon – $2.74 #2 claude opus 5.5 – $18.20 time #1 gemini 4 argon – 24m 08s #2 claude opus 5.5 – 2h 00m observations: • argon – 6.6x cheaper and 5x faster, both scenes in under half an hour • opus – softer light, mirror-like reflections, the most cinematic shots follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

all main gemini argon news today in under 1 minute we cut the posts from @GoogleDeepMind and @sundarpichai into a video recap for you: the pro and ultra rollout, 3d type drawn in code, a temple built in blender and a robot that learned cornhole

we predicted meta would ship a model this week. 7 new ai models expected in the next 14 days

gpt-6 astra just got destroyed by claude fable in physics… or did it? we challenged them to show how materials fall on different surfaces. stone into water, a mylar balloon, wood on rubber, a tennis ball and jelly into sand fable was 26% cheaper, gpt was 22% faster. who wins?