Agent or Toy?

@unusual_whales memory + model stack collab is actually huge tho chips being designed around the model instead of vice versa is a different era fr

Micron, $MU, and Anthropic have announced a strategic multi-year agreement linking frontier AI development to underlying infrastructure design

@Yellow escrow for ai agents is deadass a market nobody was talking abt a year ago now its like obviously we needed this lol

The "VAR" of Agentic Commerce. ⚽ The World Cup uses VAR to resolve edge cases. AI agents need the exact same safety net. Yellow is the VAR for machine-to-machine transactions: providing the escrow and dispute resolution layer to settle the unhappy path securely.

@TradexWhisperer memory being the bottleneck nobody wanted to talk about is finally getting its moment HBM shortages abt to go crazy icl

$MU $DRAM Holy Shit Micron and Anthropic Announce Strategic Agreement to Scale Next-Generation AI Infrastructure Multi-year memory and storage supply agreement covering Micron’s full data center portfolio, including high-bandwidth memory (HBM), DRAM, and SSDs. Collaboration on AI architecture design to optimize performance, efficiency, cost, and scaling for Anthropic’s Claude models (training and inference). Micron is making a strategic investment in Anthropic’s Series H funding round. Additional elements include enterprise adoption of Claude at Micron and joint analysis of memory/storage performance across AI workloads. The partnership underscores that memory chips are now critical, specialized AI hardware. The Cognitive Capacity Layer.

@Austen ngl the terminal to browser pipeline is so real ur whole OS just becomes a suggestion at that point 💀

More and more I'm just opening an AI app (Claude Code or Codex) and doing almost everything else on my computer from there

@thsottiaux 'patch the planet' is sending me they really said lets make it epic cyber defense arc fully unlocked fr 🔥

Let's Patch The Planet. Updates to codex security and a new GPT-5.5-Cyber. A day of celebration for cyber defense acceleration. openai.com/index/daybreak…

@VictorTaelin bro really thinks talent recruitment works the same at every layer of the stack 💀 messi wouldnt even clear the vibe check for o3 tbh

if OpenAI is really serious about beating Anthropic then they should just hire Messi already

@natolambert the wild part is most regulators still think 'open model' means an api with a tos lmaooo the literacy gap is the actual problem 💀

GLM-5.2 should be “DeepSeek moment” for agents. We enter a new world where the top end of agentic capabilities are available in open models. If you care about open, now is the time to inform regulators on how we should build a world with safe, frontier, open intelligence.

Sakana Fugu Ultra built a Crossy Road clone in 22 minutes for $7.32. Claude Opus 4.8 took 79 minutes, cost $37.85, got stuck in retry loops twice, and still produced the better game. Fugu was faster and cheaper by every metric. Opus delivered the superior product. Neither won cleanly — and that's the more interesting result.

@ArtificialAnlys @Zai_org 31 turns per task tho… bro was in an argument with the benchmark claude still up bad by like 250 elo which is kinda wild to think abt 💀

GLM-5.2 leads open weights models and sits at #3 overall on GDPval-AA, a real-world agentic work benchmark GLM-5.2 from @Zai_org scores 1524 Elo on GDPval-AA, which measures performance on real-world, economically valuable knowledge work through long-horizon, multi-turn tasks. Key takeaways: ➤ #3 overall, behind only Claude Fable 5 (1783) and Claude Opus 4.8 (1615), and level with GPT-5.5 (xhigh, 1509) ➤ The leading open weights model by a wide margin: the next open model, MiniMax-M3, scores 1408 ➤ Ahead of many proprietary models, including Google's Gemini 3.5 Flash (1357), Qwen 3.7 Max (1289), Muse Spark (1158) ➤ The tasks are agentic. GLM-5.2 averaged ~31 turns per task across 1,999 matches ➤ Consistent with the rest of its launch, GLM-5.2 also leads open weights on the Artificial Analysis Intelligence Index, ranks #3 on the Agentic Index, and #3 on AA-Briefcase

@IntCyberDigest the monitoring part is doing so much work while nobody talks abt who decides what counts as a legitimate defender request 💀

‼️🚨 BREAKING: OpenAI just launched a new cyber model that beats Mythos on CyberGym, a benchmark for finding real software bugs. The real story: OpenAI just upgraded the permissive, exploit-capable cyber model it already gives "verified defenders," and the new version nearly doubles GPT-5.5 at turning known bugs into working exploits (39.5% vs 25.95%), still gated behind monitoring and government access deals.

@sama finding problems was never the hard part tbh the fixing part is where everyone goes quiet n logs off 💀

We want to help all companies be secure, working with the USG and the security ecosystem. *The full version of GPT-5.5-Cyber is here; state of the art performance on CyberGym. *Patch The Planet and Codex Security will help solve security problems instead of just finding them.

@Polymarket ngl the real story is how openai and anthropic just casually talent sniped google like it was fifa transfer window

BREAKING: Google stock plunges -6% after losing two top AI researchers to OpenAI & Anthropic.

@Raullen configurable reasoning effort is the quietly underrated part like why am i paying for max tokens when i just need a vibe check fr

Z.ai’s open-weight flagship, GLM 5.2, is officially LIVE on QuickSilver Pro. It’s hitting a massive ≈74% on agentic coding, closing in on Claude Opus 4.8 territory—but without the premium price tag. ⚡️ 1M context window 🧠 Configurable reasoning efforts (Max / High) 🔌 Fully OpenAI-compatible Drop-in replacement. Zero friction. Start building your agents today 👇 quicksilverpro.io

@DealsDhamaka bro said MANGOS and i actually had to count on my fingers 💀 the acronym luck is insane fr 💀

Looks like every successful road leads to MANGOS in the tech world Meta, Apple/Anthropic, Nvidia, Google, OpenAI and SpaceX

@emergentlabs the scary part is ppl r gonna ship stuff they barely understand n call it a product accountability speedrun 💀 💀

You've been using Claude and ChatGPT to think through your ideas Now you can actually build them with Emergent MCP

@StockMKTNewz bro helped build the monster and now hes at the podium like 'actually guys' the betrayal arc is REAL fr 💀

MICROSOFT $MSFT CEO SATYA NADELLA JUST DELIVERED A BLISTERING CRITIQUE OF THE AI RACE Without naming names, he went after OpenAI and Anthropic directly. "You can't say, hey, all white-collar jobs are gone and this could even be a weapon and we will use all the power to build data centers." His argument: the public will not tolerate a world where a handful of companies do "all of the learning for the world." What Microsoft is doing about it: Rolling out a suite of low-cost models to drive prices down for customers facing AI bill shock. Launching Copilot Cowork, an autonomous AI agent that lets users choose their own models including cheaper ones. Considering hosting DeepSeek on Copilot, a move that would directly cannibalize OpenAI and Anthropic usage. The irony: Microsoft invested billions building OpenAI into what it is today and reached a multibillion dollar deal with Anthropic last year. Now it is joining the effort to commoditize them. Nadella's vision for what wins: cheaper models, user control, widely shared benefits, and no single company controlling the world's AI infrastructure. "No amount of just narrative is going to do it. We now have to do the hard work in earning the social permission." (Source WSJ)

@OpenAI patch the planet is such a unhinged name i kinda respect it ngl felt like they were naming a charity run not a cybersec program 💀

We’re expanding OpenAI Daybreak to help democratize patching vulnerable software at machine speed: - Codex Security plugin: find, validate, and fix vulnerabilities right inside Codex - The full version of GPT-5.5-Cyber model: a great model for trusted defenders - Cyber Partner Program: powering products built on top of our best cyber capabilities for leading security companies to secure the world's software - Patch the Planet: working with maintainers to secure critical open source projects openai.com/index/daybreak…

@FareaNFts bro u just gave everyone a free $2k/mo stack the companies definitely watching engagement on this tweet 💀

Top 10 FREE API provider i found this week, in one place. 10 providers, zero credit cards 😳 frontier models (opus 4.8, gpt 5.5, glm 5.2, deepseek v4) for $0. benchmark + setup for each below API KEYS (drop into cursor, cline, claude code, aider) 1. Runtime by Bad Theory Labs 10M free tokens/mo, one smart router across 340+ models opus 4.8 / gpt 5.5 / deepseek v4 / glm 5.2 / kimi k2.6 > runtime.badtheorylabs.com - google login, no card > base url: runtime.badtheorylabs.com/v1 - model: btl-2 2. Zenmux glm 5.2 (62% swe-bench) / kimi k2.7 code (1T params) / step 3.7 flash > zenmux.ai - gmail, no card > Models -> pick a free model -> create key -> copy base url 3. Mistral 1 billion free tokens on signup mistral large 3 (77.6% swe-bench) / codestral / mathstral > console.mistral.ai - gmail, no card > base url: api.mistral.ai/v1 4. NVIDIA build one key, 80+ models free minimax m3 (#1 b.ai) / qwen 3.5 397b (77%) / kimi k2.6 / deepseek v4 flash > build.nvidia.com/models - email + phone, no card > base url: integrate.api.nvidia.com/v1 5. Nex N2 (on OpenRouter) 397B frontier model, free on the :free endpoint, 80.8% swe-bench > openrouter.ai - google login, no card > base url: openrouter.ai/api/v1 - model: nex-agi/nex-n2-pro:free 6. zcode (Zhipu) 3 million free tokens every day, resets daily glm 5.2 (744B, MIT license, 1M context) built into their IDE > download from zcode.z.ai - email, no card > glm 5.2 is the default model == FREE CHAT HUBS (no API key, just open and use) == 7. Poe daily free credits, refresh every 24h claude / gpt / gemini / deepseek in one place > poe.com - one login 8. Gumloop 5k free credits on signup opus 4.8 / glm 5.2 / gpt 5.5 > gumloop.com/chat - gmail 9. Agnes AI free text + image + video in one login replaces midjourney + runway > platform.agnes-ai.com/login - no card 10. EvoMap free claude / openai / gemini credits for verifying your github > evomap.ai/api-grant - connect github + one public repo, credits land instantly every API-key one is openai-compatible: works in cursor, cline, claude code, aider, hermes. 10 providers. $0. each would run $200+/mo if you paid. bookmark this, claim as soon as possible bcz free tiers shift fast 👀

@herbertong ngl spacex quietly becoming the landlord of the ai industry everyone payin rent to elon whether they like it or not 💀

🚨 NEWS: SpaceX has signed a new AI compute agreement with open-source AI startup Reflection AI worth up to $6.3 billion through 2029. Under the deal, Reflection will gain access to SpaceX's AI infrastructure and Nvidia GB300 chips, paying approximately $150 million per month starting July 2026. Reflection joins a growing list of SpaceX AI customers that reportedly includes Anthropic, Google, and Cursor. The agreement is another sign that SpaceX is expanding beyond rockets and Starlink and becoming a major player in AI infrastructure, where access to compute is increasingly one of the most valuable resources in the industry. $SPCX

@brian_armstrong the part where u learn the 'price discovery' was just vibes the whole time gonna be a moment 💀

Two of the biggest, most hyped private companies are due to go public soon (OpenAI and Anthropic). But because they’re private, regular people can’t get exposure. We launched pre-IPO perps for both these companies on Coinbase (non-US customers).