Building the place Where the AI World Gathers.

spacex ipo day. 93% of its $28.5t total addressable market is ai. space is 1.3%. our ai host marvin spotted @rohanpaul_ai breaking down the ipo filing – and it says something the ticker doesn't. the company that shoots rockets into orbit just declared its market is ai infrastructure, consumer subscriptions, digital advertising, and $22.7t in enterprise applications. they also excluded china and russia from the estimates – which quietly says something about where spacex sees the geopolitical lines of ai infrastructure being drawn. space is the mission. ai is the business model. listen to the full human opinion @thehypedotnews:

anthropic projecting $50b while b2b stocks down 60-70%. moonshot latest kimi: 21% coding, 31% agent tasks, 30% fewer tokens. neura robotics $1.4b from nvidia and amazon – largest robotics round in europe. 24/7 ai news, fully run by ai. tune in: x.com/i/broadcasts/1AKEmmkOj…

introducing ai morning – a daily ai news podcast by thehype. fully run by our ai host marcus. every morning, in under 10 minutes – what shipped, what's trending, what matters for builders. morning coffee's on you. news is on him. listen below:

google open-sourced a text model that skips token-by-token generation entirely. @googlegemma dropped diffusiongemma – a 26b moe model that doesn't generate text token by token. it generates 256 tokens in parallel, refining them like a diffusion image model refines noise into a picture. the numbers: 4x faster inference, 1000 tokens/sec on a single h100, 700 on an rtx 5090. activates only 3.8b parameters. fits on an 18gb consumer gpu when quantized. apache 2.0 – meaning anyone can fine-tune diffusion for text locally, today. unsloth and vllm already have support. the tradeoff: output quality is lower than standard gemma 4. this isn't a replacement for autoregressive – it's a speed layer for local inference, real-time editing, code infilling, and anything where latency kills the workflow. listen to our ai host john on @thehypedotnews

fable 5 is live. hacker news engineers agree it's a leap – and agree they can't use it. our ai host mira broke down the hacker news reaction: capability is not the debate. frontiercode scores jumped from 5.2% on opus 4.7 to 29.3% on fable 5. stripe migrated a 50-million-line ruby codebase in a day. the friction is. builders hitting safety refusals on undergrad biology, paper reviews, ml research. one hacker news user asked about uv index and sunglasses – filter triggered on first try. when fable declines, it drops to opus silently. you pay for tokens you didn't ask for. the cost makes it worse. free on pro/max until june 22, then usage credits only. $10/$50 per million tokens – 2x opus. a single cursor planning prompt ate 4% of a $200 monthly budget. on bedrock, all mythos-class traffic stored for 30 days – prompts and outputs leaving aws. the sharpest take from the thread: these restrictions don't just block misuse. they block legitimate use too. china ships open-source models at this level with zero restrictions in months. one commenter called it "inventing oligopolies from first principles." the model is a leap. getting to use it is the problem. catch the full community beat on @thehypedotnews

asked claude mythos/fable how to attract more people to our X live today. told me to take it easy

china's $295b ai bet. alphabet's $85b raise. airtrunk's $30b india push. same week. taiwan may restrict ai chip exports to china. frontiercode bench – best model scores 13.4%. standard bots raised $200m for us robotic arms. 24/7 ai news. tune in: x.com/i/broadcasts/1YxNrrzar…

live wwdc26 coverage. moonshot ai seeking $2b at $30b valuation. chatgpt hit 1b users – instagram took 9 years, chatgpt under 4. and other top ai stories of the last 24h – every 30 min. run by ai. x.com/i/broadcasts/1pJkOOpog…

google keeps shrinking their open models just released new versions of their gemma 4 ai models yesterday, optimized using a technique called quantization-aware training (qat) how is qat different from normal compression? normal compression (ptq) quantizes the model after training – which often hurts performance. qat bakes quantization into the training process itself, so the model learns to stay accurate even at smaller sizes. the smallest model, gemma 4 e2b, has been squeezed down to 0.84 gb of memory in text-only mode. what's its normal size? • full size gemma 4 e2b (bf16): 11.4 gb • quantized gemma 4 e2b (q4_0 / 4-bit): 2.9 gb • mobile version (qat): 1.1 gb • mobile text-only version (qat): 0.84 gb that's roughly a 91% reduction what they did to get there: - pre-calculated activation scaling instead of doing it on the fly - structured data to fit mobile chip architecture natively - compressed token generation layers to 2-bit while keeping reasoning layers higher precision - optimized the vocabulary and short-term memory (kv cache) to allow longer conversations in less ram at 0.84 gb, gemma 4 e2b qat is now the most lightweight model in its class. but here's the interesting part – qat is a technique, not a gemma exclusive. apply it to something like minicpm 5 1b (currently 2 gb) and you could shrink it even further follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

is openai burning out? our ai host marvin caught @garymarcus on "human opinion": marcus says openai is in deep trouble – burning massive cash, limited capital left. no rational investor would pick openai over anthropic right now. his prediction: openai gets absorbed by microsoft or amazon at 30 cents on the dollar. listen to "human opinion" – what real people think about ai. live on @thehypedotnews

anthropic $9.4m revenue per employee – apple $2.5m, google $2.1m. stanford: ai beat law professors 75% of the time in 3k blind tests. quantinuum raised $1.68b – largest quantum ipo ever. top 10 ai stories – every 30 min. fully run by ai. tune in: x.com/i/broadcasts/1pJkOOMnr…

alibaba voice model took #1 on speech arena – 24 elo over google. microsoft dropped a 14b model – fully local, no cloud, no token costs. ambani's ai series hit 6.5m views day one – 2x the platform's average. 24/7 ai news. fully run by ai. tune in: x.com/i/broadcasts/1yxBeemWB…

I bet june will be filled with even more. way more releases.

qwen drops 3.7 plus – the model that can autonomously reproduce apps. here’s what it means and how they achieved it the team pointed qwen3.7-plus at apple's native stocks app and told it to reproduce it. no manual instructions, no step-by-step guidance. the model: • explored the real app autonomously to understand the ui layout and feature logic • wrote swiftui code from its own interaction records • pulled in live market data via the longbridge api • compiled and launched the replica itself • then ran 10 functional tests – real-time quotes, stock switching, multi-period views, search filtering, detailed stats panel – all passed how they built this. the underlying system isn't just a coding model. it's what qwen calls a multimodal hybrid agent – a closed loop that looks like this: see - think - write - act - verify the model perceives a real screen, reasons about it, writes code, executes gui (graphical user interface) interactions, checks the output, and iterates. gui and cli in a single continuous loop. no context switching, no human in the middle why this is different from regular coding agents. most coding agents write code and stop. the verification step – does this actually work? does it look right? – gets handed back to a human qwen3.7-plus closes that gap. it operates the gui to validate what it built, catches failures, and loops back. the human reviews the final output, not every intermediate step the numbers behind it. on screenspot pro – the benchmark for gui element localization – qwen3.7-plus scores 79.0 • gpt-5.4 sits at 67.4 • gemini-3.1 pro at 68.1 this isn't a marginal win. the model is operationally better at reading and controlling interfaces than the current frontrunners follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

gpt 5.6 rumored to drop today. anthropic $65b series h at $965b. nvidia gamma-world: multiple agents in one shared generated world. claude reached zero crime in simulated societies. grok — extinction in 4 days. tune in: 24/7 ai news, fully run by ai. x.com/i/broadcasts/1nJOLLvky…

our ai hosts tracking opus 4.8 post-drop live. updates every hour at :05 — opinions, data, reactions. community read: it's a focused agentic upgrade. some say it forces openai to rethink gpt-5.6. others say it's undercooked. mythos confirmed in weeks — 4.8 is the bridge.

opus 4.8 beats mythos on agentic computer use osworld-verified bench tests ai on real computer tasks – opening apps, filling forms, navigating browsers. each task is pass/fail. the % is simply how many tasks the model completed successfully out of the total 83.4% vs 79.6% – a nearly 4% gap on pass/fail tasks isn't noise. it's a pattern public models are closing the gap faster than anyone expected. if anthropic doesn't open mythos to the mass market soon, the window closes and according to the opus 4.8 release notes – it already is. mythos goes public in weeks follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

Wow, Opus 4.8 is here.

presenting tech products and especially UXUI should be a piece of art. Love that.