Daily Intelligence Briefing: Tech / AI / IT Monitor

Date: 2026-06-17

---

Executive Summary

GLM-5.2 from Zhipu AI (@Zai_org) emerged as the dominant story of the day, launching as a new flagship open-weights model optimized for 1M-token context and long-horizon agent tasks, ranking #2 in Code Arena for frontend (behind only Fable-5). Cursor's explosive commercial traction was confirmed—enterprise revenue run rate grew 50x YoY, with reports valuing the VS Code fork at $60B and generating $4B in revenue. On the hardware front, benchmarks revealed that the community-built Vulkan driver outperforms AMD's official ROCm stack by 17% on token generation for local AI inference on AMD's Strix Halo APU, raising serious questions about AMD's software investment. Meanwhile, inference reliability remains fragile, with providers buckling under traffic from new model launches.

---

Key Events

• GLM-5.2 launches as top-tier open-weights model: Zhipu AI released GLM-5.2 with 1M-token context, optimized for long-horizon agent coding tasks. It ranks #2 in Code Arena: Frontend, trailing only Fable-5 (which is unavailable/closed). Day-0 support added in Ollama, Exo 0.14.0, and Codex. Multiple testers reported superior performance over Fable-5 for CUDA/WebGPU kernel generation. → link

• Cursor's enterprise revenue run rate grew 50x year-on-year, with reports surfacing that the VS Code fork sold at a $60B valuation with ~$4B in revenue. → link

• Vulkan (community driver) beats AMD's official ROCm by 17% on token generation for local AI inference on AMD Strix Halo (128GB). Same model, same chip—only the driver differed. → link

• Jack Dorsey signals new intelligence tools initiative: "we're going to start talking a lot more about our intelligence tools. this is the beginning of the beginning." → link

• GPT-Realtime-2 teased by Greg Brockman, described as "something new" with no further details. → link

• Inference infrastructure remains unstable: "every time we roll out a new model our providers go down (GLM 5.2 right now) with the traffic. Gonna be a while before these things are as reliable as traditional infra." → link

• Hermes Agent updates: Teknium shared a PR adding API key security (making keys inaccessible to agents), and changed the Curator to only prune unused skills by default—saving some users $16+/week. → link

• Epic Games launches version control system positioned as a Perforce/SVN alternative, generating significant developer community buzz. → link

• Bland raises $100M Series C to automate complex phone calls. → link

• Postgres reported to run half as fast on Linux 7.0, with known root cause identified. → link

---

Analysis

Patterns observed: - The open-weights ecosystem is accelerating rapidly. GLM-5.2's immediate Day-0 integrations (Ollama, Exo, Codex, MLX forthcoming) show a mature release pipeline that contrasts with the gated/unavailable Fable-5. The community is vocal about wanting maximum capability in open weights regardless of parameter count (10T was explicitly mentioned). - AI coding tool "wrapper wars" are intensifying—thdxr noted that even tools for building other tools are getting wrapped with another access layer, calling it "full of anxiety, zero restraint." This suggests market saturation and impending consolidation. - Local AI inference on non-NVIDIA hardware is viable but bottlenecked by vendor software, not silicon. AMD's hardware competes; its software stack doesn't. This is a recurring pattern (cf. Epyc's trajectory). - Cursor's $60B valuation and 50x enterprise growth signal that developer tools with AI integration are the current premium acquisition targets—SPCX/Cursor acquisition chatter reinforces this.

Escalation indicators: - GLM-5.2 benchmark results place open-weights models within striking distance of closed proprietary leaders in coding tasks. - Multiple independent testers (Ex0byt, louszbd, sam_paech) confirming GLM-5.2's strength suggests genuine capability, not marketing.

What to watch next: - Qwen-3.7-27B was teased but not yet released—arrival imminent. - OpenAI's "major leap" voice mode upgrade for ChatGPT reportedly coming soon. - Fable-5's response to GLM-5.2 encroaching on its #1 Code Arena position. - AMD's response to the Vulkan > ROCm benchmark gap—whether they close it or lean into community drivers. - Jack Dorsey's "intelligence tools" announcement specifics.

---

Tweet Feed

AI Model Releases & Benchmarks

@Ex0byt · 2026-06-16T23:23

Impressive. This is what a top open source model looks like. GLM-5.2 is stable at 1M token context tasks purposefully optimized for long-horizon agent coding work trailing Fable-5 @ ~3x flops efficiency. → tweet link

@Ex0byt · 2026-06-17T15:09

GLM 5.2-PRISM-Pro not refusing to write CUDA kernels for my Blackwell (or, for that matter, any other code I tell it to write) is priceless. I can't wait to share this gem with y'all. → tweet link

@Ex0byt · 2026-06-17T14:02

Yeah.. much better than Fable-5 for generating WebGPU kernels/shaders - GLM-5.2 - absolutely stoked! → tweet link

@Ex0byt · 2026-06-17T13:23

GLM-5.2 testing testing 1..2..3.. → tweet link

@Ex0byt · 2026-06-17T13:00

more validations.. I have this beast running right now and I am just blown away.🤯 → tweet link

@Ex0byt · 2026-06-17T18:38

HaaaaaaydenH, Alibaba_Qwen Qwen-3.7-27B 🤗 wen? → tweet link

@louszbd · 2026-06-16T21:35

glm-5.2 is built around 1M token context window. on long-horizon tasks, it stays more focused. on top of that, you can get a noticeably better experience it brings at the same price as before. after some tests, we felt the more we use it, the smarter it feels... → tweet link

@TheAhmadOsman · 2026-06-16T23:09

GLM 5.2 numbers make me believe I was too conservative in my own prediction. 2 months tops and we'll have Fable 5 at home → tweet link

@TheAhmadOsman · 2026-06-17T08:51

GLM 5.2 → tweet link

@ollama · 2026-06-17T18:03

You can use GLM-5.2 and Kimi-K2.7-Code in Codex with Ollama! ollama launch codex / ollama launch codex-app → tweet link

@gdb · 2026-06-17T04:23

GPT-Realtime-2 is something new → tweet link

@RydMike · 2026-06-17T17:42

Well, the first two were just teasers, the app actually dropped though. Still I'm looking forward to using the new model when it drops, if it remains price performance competitive → tweet link

@alexocheema · 2026-06-16T21:52

MIT-licensed open weights 💛 → tweet link

@victormustar · 2026-06-17T10:18

sorry but the "I cannot run this model it's 800B params" well we don't care that's not what's at play here. We want 10T open weights if possible. We want maximum capability that is open :) → tweet link

@victormustar · 2026-06-16T19:57

and it will cost x30 less 👀 → tweet link

Hardware & Infrastructure

@sudoingX · 2026-06-17T13:06

the fastest way to run local ai on an amd chip, it turns out, is to not use amd's software. [...] rocm, amd's official stack: pp 947, tg 45.6 tok/s. vulkan, the volunteer-built driver: pp 957, tg 53.6 tok/s. [...] token generation, the free community driver is 17 percent faster. on amd's own silicon. than amd's own software. → tweet link

@sudoingX · 2026-06-16T19:58

slowly? i just put it on the bench. [...] a first-gen amd apu [...] came within 10% of a purpose built nvidia ai box on token generation, the speed you actually feel, at a third of the price. and it loaded a 397 billion parameter model on a desk. [...] the people sleeping on amd right now are the same ones who slept on epyc in 2018. → tweet link

@sudoingX · 2026-06-16T19:36

yep, vulkan beat rocm by ~17% on token gen, dead tied on prompt processing. [...] so the gap held WITH the tuning, not for lack of it. → tweet link

@TheAhmadOsman · 2026-06-17T04:10

Memory prices are crazy stupid ngl → tweet link

@gospaceport · 2026-06-17T01:47

Quad 3090s, just beautiful 🔥 → tweet link

@gospaceport · 2026-06-17T01:27

It kinda works awesome in Proxmox also → tweet link

@hnasr · 2026-06-17T15:39

Postgres is half as fast in Linux 7.0, and we always knew why → tweet link

@thdxr · 2026-06-17T15:40

inference is still so primitive. every time we roll out a new model our providers go down (glm 5.2 right now) with the traffic. gonna be a while before these things are as reliable as traditional infra → tweet link

AI Agents & Frameworks

@Teknium · 2026-06-17T17:06

Had my Hermes Agent evaluate headroom. Key conclusions: Most of what it would do for Hermes actually is net greater token cost 🧐🧐 There was one aspect it discovered would save ~60% of tokens used on the search file tool, which we will be integrating! → tweet link

@Teknium · 2026-06-17T14:26

Would love more eyes on this PR that adds security to all your API keys and makes them inaccessible by the agent. If you have a security focused background, please post any reviews or concerns! → tweet link

@Teknium · 2026-06-17T12:28

Making a change to the Hermes Agent Curator that will make it so it only Prunes unused skills by default, it will no longer consolidate skills unless you opt in. It was costing upwards of 16$ per week to operate for some using frontier models and causing some confusion. → tweet link

@Teknium · 2026-06-17T13:27

Tip of the Day: Did you know you can share your hermes agent profiles onto github? Share a profile and others or yourself can download and distribute it with ease! → tweet link

@TheAhmadOsman · 2026-06-17T11:43

Hooked up Cloudflare and Porkbun APIs into an agent and that's how I purchase domains, fixup DNS, and deploy everything through Cloudflare now. This busywork used to take so much time to configure and now it is done in the background while I am doing more important stuff → tweet link

@LinusEkenstam · 2026-06-17T14:49

I typed one sentence. I got back 5 production-grade prompts. One each for Midjourney, Nano Banana Pro, ChatGPT Image 2, Seedance 2 and Kling Omni. Then I shipped the app behind it in a day. → tweet link

@MengTo · 2026-06-17T13:47

Framer 3.0 just came out and I might actually start using it. - Much easier to start thanks to Agents - Creates perfect components, auto layout and even breakpoints - World-class ecosystem of templates and components. But holy crap, is it hungry for tokens. → tweet link

@nummanali · 2026-06-17T11:42

Omg this actually is the best production agent framework for anyone to use. Literally does everything and you worry about zero deployment issues. Needs Vercel for the best out of box experience. But given it's OSS, more platforms will become available hopefully → tweet link

@cooltechtipz · 2026-06-17T00:36

Guide to AI agents and workflow automation. → tweet link

Developer Tools & IDEs

@ASalvadorini · 2026-06-17T07:57

Bye bye Cursor. It has been nice while it lasted 👋 #AI #cursor → tweet link

@MatejKnopp · 2026-06-16T20:42

I had no idea that cursor had 4B revenue. What the fuck. I think they have been popping champagne long time ago. → tweet link

@thdxr · 2026-06-17T14:48

this dynamic is really crazy right now. even things for building other things are getting wrapped with another layer trying to be the access point. everyone is flailing, full of anxiety, zero restraint → tweet link

@steipete · 2026-06-17T02:59

RT @btraut: Codex now supports .worktreeinclude! Worktree enjoyers can now automatically copy over things like .env files, config files, agent… → tweet link

@jsuarez · 2026-06-17T15:44

[Long thread on PufferLib 4.0 algo advancements and Codex usage] The quality of the code you get out of Codex is what you'd expect from a particularly stupid undergrad. [...] Still aiming for end of summer for 5.0 → tweet link

@gdb · 2026-06-17T00:36

software engineering is so different now. hard to remember what it was like even 6 months ago. → tweet link

@cooltechtipz · 2026-06-17T12:53

The most valuable skill in tech right now might be taste. AI can write code. AI can write content. AI can make designs. But somebody still has to decide what's actually good. The bottleneck is slowly moving from creation to judgment. → tweet link

Open Source & Research

@jezell · 2026-06-17T15:49

RT @turbopuffer: we open sourced alyze, the Rust crate behind tpuf's default full-text search tokenizer (word_v4) → tweet link

@jsuarez · 2026-06-16T21:25

6,000 stars! Thank you all! We're puffing up our core RL algo in the next update, targeting end of summer! → tweet link

@cooltechtipz · 2026-06-17T16:36

Evolution of sequence models in deep learning. → tweet link

@cooltechtipz · 2026-06-17T10:13

How neural networks learn. → tweet link

@cooltechtipz · 2026-06-17T06:32

AI ecosystem → tweet link

@badlogicgames · 2026-06-16T19:17

RT @gregbarbosa: Apple didn't, so I did: I made it dead simple to run macOS 27's local and Private Cloud Compute Foundation models in any app… → tweet link

@badlogicgames · 2026-06-17T12:27

RT @mitchellh: We've gone really quickly from "local models are dogshit" to "local models are good actually" (like, a 12 month window from… → tweet link

Tech Industry & Business

@TrungTPhan · 2026-06-17T19:00

RT @bearlyai: Cursor's "revenue run rate from enterprise customers grew 50x year-on-year" → tweet link

@jezell · 2026-06-16T19:55

RT @LLMJunky: JUST IN: VS Code fork sells for 60 Bs. With a capital B. And an uppercase 60. That's a lot of cash. 🤯 Cursor is amazing, h… → tweet link

@levelsio · 2026-06-17T17:16

Stripe would be an obvious buy for SpaceX I think. Instantly get X Money to work and become WeChat (which Elon Musk said he wanted to become the "everything app"). He also seed invested in Stripe. It'd be $159B → tweet link

@FinansowyUmysl · 2026-06-17T07:06

SPCX przebił 200$. Ci którzy kupili go w dniu debiutu - zarobili już 48%. Jednocześnie SPCX wykorzystuje swoją kapitalizacje i robi zakupy na rynku - tym razem Cursor. Po co? Ogromna baza użytkowników, na której można trenować Groka. → tweet link

@LinusEkenstam · 2026-06-17T07:30

RT @usebland: Today, we're announcing that we've closed our Series C and raised $100M to continue automating the world's most complex phone… → tweet link

@jack · 2026-06-17T18:53

we're going to start talking a lot more about our intelligence tools. this is the beginning of the beginning. → tweet link

@levelsio · 2026-06-17T03:09

Absolute pleasure to finally meet @rrhoover at @cursor_ai Compile today → tweet link

Apple & On-Device AI

@TrungTPhan · 2026-06-16T22:35

AirPods with camera aren't for taking photo or video. But will be used to give visual intelligence to Siri. A recent Apple navigation patent shows use case: instead of saying "turn right in 100m", the AirPod camera can see and provide natural-sounding directions such as "turn left after the coffee shop" → tweet link

@TrungTPhan · 2026-06-17T16:17

The Apple Maps turnaround is impressive. [...] now spent billions mapping >4 million miles of proprietary street level maps data with specialized vans and people wearing goofy Apple backpack cameras [...] using its powerful custom silicon to give AirPods visual intelligence → tweet link

@jezell · 2026-06-16T23:38

RT @testingcatalog: OPENAI 🔥: More details about the upcoming voice mode upgrade for ChatGPT. > It will be advertised as a "major leap in… → tweet link

@FrameworkPuter · 2026-06-17T16:30

RT @tommybee456: I'm making another @FrameworkPuter gaming handheld, but this time, it's modular! → tweet link