Executive Summary
OpenAI’s GPT-5.6 Sol model is experiencing explosive growth, reaching 8 million active users across Codex and ChatGPT Work, driving intense demand on inference infrastructure. On the open-source frontier, the 1-trillion parameter "Inkling" model (41B active) was released by Thinking Machines, offering advanced reasoning across text, image, and audio. Meanwhile, developers are rapidly adopting WebGPU and WebAssembly to build cross-platform applications, and the local AI community is gaining momentum, advocating for self-hosted models due to cost-efficiency and data sovereignty.
Key Events
- GPT-5.6 Sol Explosive Growth: OpenAI's GPT-5.6 Sol hits 8M active users, causing the inference team to "move mountains" to scale capacity. → link
- Inkling 1T Parameter Model Released: Thinking Machines introduces Inkling, an Apache-2.0 open model with ~1T parameters (41B active) capable of processing text, image, and audio. → link
- GPT-Red Launched: OpenAI introduces an internal automated red teamer designed to find prompt injection vulnerabilities at scale. → link
- GLM 5.2 Performance Milestones: LMSYS and SGLang reach 500 TPS on GLM5.2 NVFP4 Agentic Workload using 8xB300 hardware, while tinygrad demonstrates the feasibility of running Opus-level models at 100 tok/s on CPU-GPU hybrids. → link
- WebGPU Advancements in Flutter: @jezell showcases a custom Flutter engine running on Skia Graphite with WebGPU bindings, enabling buttery-smooth WASM applications. → link
- Hermes Agent Updates: Teknium's Hermes Agent adds credential pooling for multiple accounts, parallel tool call optimizations, and integrates the Blender MCP. → link
Analysis
The AI landscape is bifurcating between massive proprietary API models (like GPT-5.6 Sol) and highly capable open-weight models (like Inkling and GLM 5.2). While proprietary models are capturing the bulk of immediate developer mindshare for coding and business automation, there is a growing underlying anxiety about open-weight model bans and vendor lock-in, driving a parallel surge in local AI tooling (Ollama, OpenCode). On the software engineering side, WebGPU and WASM are emerging as the definitive stack for high-performance, cross-platform web and desktop rendering, bypassing traditional engine bottlenecks.
Tweet Feed
AI Models & Research
@gdb · 2026-07-15T18:45
GPT-Red — improving model security through automated red teaming of prompt injection vulnerabilities: → tweet link
@victormustar · 2026-07-15T18:55
RT @LysandreJik: Thinking Machines' Inkling is out: first ever open and large (1T), text, image and audio in, text out. One thing I find q… → tweet link
@victormustar · 2026-07-15T18:35
RT @natolambert: Thinky with a ~1T param, 41B active, apache-2 model. Benchmarks are a clear step up from Nemotron Ultra (55B active), new b… → tweet link
@victormustar · 2026-07-15T17:41
Grok 4.5 Boeing Benchmark result: probably somewhere between Opus 4.5 and 4.8 imo (excited for Grok 5) https://t.co/bsMOnX9XVc → tweet link
@kunchenguid · 2026-07-15T17:41
after a few more days of using gpt 5.6 sol, i started noticing some issues - if you have good solutions, please share! 1. it uses technical jargons a lot... 2. it can over-engineer and spiral out of control... 3. it's overly conservative in terms of touching live environment... → tweet link
@tinygrad · 2026-07-15T18:36
Once DDR5 prices come back to earth, some reasonably priced CPU-GPU hybrid machines for GLM 5.2 at 100 tok/s look feasible. What would you be willing to pay for an Opus level model in your living room? → tweet link
@tinygrad · 2026-07-15T18:15
Looks like DRAM read bandwidth on EPYC 9334 is limited by the CCD -> I/O die links, not by the RAM itself. Anyone figure out how to beat 309 GB/s? https://t.co/YR0hi91eSI → tweet link
@TheAhmadOsman · 2026-07-15T06:36
Kimi K3 is glorious → tweet link
@TheAhmadOsman · 2026-07-15T00:55
Know why Anthropic hates Opensource AI? GLM 5.2 being free and available to download made their $1 Trillion valuation make no sense → tweet link
@TheAhmadOsman · 2026-07-14T21:23
Kimi K3 is almost done cooking. We're gonna be eating good → tweet link
@sama · 2026-07-14T19:02
5.6 sol growth is insane. the inference team has done heroic work to be able to support demand. we are going to move mountains to continue to scale, but it is possible there are some hiccups soon. → tweet link
@louszbd · 2026-07-15T03:26
RT @lmsysorg: Serving GLM5.2 NVFP4 Agentic Workload with SGLang: How We Reached 500 TPS on 8xB300 at bs=1. In this deep dive, we explain ho… → tweet link
@sudoingX · 2026-07-15T01:49
qwen 3.6 35b a3b census. this model runs on everything, so let's prove it. reply with your rung: cpu only, single gpu, multi gpu, or dgx cluster. add your tok/s and usable context before it drags. → tweet link
@MilksandMatcha · 2026-07-14T19:05
NYC builders: next Tuesday we’re turning @datadoghq into Cafe Compute with @GoogleDeepMind ☕️ Come try Gemma 4 31B at @cerebras speed, get increased model access, and meet the teams behind it. → tweet link
@RayFernando1337 · 2026-07-15T08:34
RT @anshuc: This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) → tweet link
Developer Tools & AI Coding
@Teknium · 2026-07-15T17:04
FYI Hermes Agent supports credential pools, which allows you to auth multiple accounts - when one runs dry it will automatically switch to the next one. Learn more on the docs: https://t.co/NbVXcKQlZN → tweet link
@Teknium · 2026-07-15T16:05
The Blender MCP is now part of the Hermes Agent MCP Catalog! Easily activate and install the Blender MCP by running
hermes mcp install blenderand ask your agent to start using blender. → tweet link
@Teknium · 2026-07-15T20:46
Prior to today, tool calls would work in parallel only if all were safe to parallelize. Now you gain big speedups for parallel tool calls so long as any subset are parallelizable. → tweet link
@Teknium · 2026-07-15T03:01
A lot of people have said gpt is unusably slow in Hermes Agent today - to the best of my knowledge there is an outage or it is overloaded. Will update if i learn or discover more → tweet link
@MengTo · 2026-07-15T02:11
Blender MCP is kind of wild. Zero 3D experience with blender skills from https://t.co/YttKmaOcEo is enough to create intricate, animated models. Here's the github repo: https://t.co/2I2rctmiRW → tweet link
@ollama · 2026-07-15T00:33
You can use OpenCode Desktop with Ollama! Try it with the top open models! https://t.co/I3VSgOCRCo → tweet link
@thdxr · 2026-07-14T21:53
you're generally working on a few sessions and now you can keep them within reach. sessions in the same project, across projects, even local and remote servers. and this will all get turbocharged when we drop 2.0 → tweet link
@thdxr · 2026-07-14T15:30
think this is the first opencode fork to be acquired → tweet link
@RayFernando1337 · 2026-07-15T14:57
RT @v_pradeilles: Don't forget: Xcode 27 comes bundled with coding skills that you can export to Claude Code or Codex by running this comma… → tweet link
@Ex0byt · 2026-07-15T14:56
do not turn on codex auto reload without an explicit max limit set. you've been warned. https://t.co/bzJG0hFkTW → tweet link
@juliarturc · 2026-07-15T06:18
I've been tweaking my workflow for making Remotion animations for over a year now... A couple of days ago I added Remotion and Hyperframe connectors to Codex. My intricate workflow is useless. I'm a simple woman now: I type what I want to see, and I see it. → tweet link
@swyx · 2026-07-14T22:43
uhm this gpt 5.6 launch might be the openai's most successful model ever since... since chatgpt? this is IPO altering stuff going on here https://t.co/c66oEVEOVF → tweet link
@swyx · 2026-07-14T21:27
RT @myprasanna: Launching @vorfluxai : The autopilot for software engineering. I was prev co-founder / CTO of @Rippling ($10B) and #1 code… → tweet link
@ivanfioravanti · 2026-07-14T20:48
MLX#LM-LoRA 3.0! 🚀🚀🚀 → tweet link
@ivanfioravanti · 2026-07-14T20:12
UnslothAI is playing in its own league! What a team! 🚀 → tweet link
@jxnlco · 2026-07-15T01:17
RT @nicoalbanese10: I’m joining @OpenAI to help build the Codex app! I’ve been using Codex all day, every day for months. It’s where I wri… → tweet link
@jxnlco · 2026-07-15T02:58
RT @OpenAIDevs: 7M+ weekly Codex users. 150+ updates in two months. @romainhuet catches you up on what’s new in Codex: GPT‑5.6 and Ultra, P… → tweet link
@jxnlco · 2026-07-15T13:57
RT @nickbaumann_: Reminder: ChatGPT for Teachers is free for verified U.S. K–12 educators through June 2027. It’s a secure workspace to pl… → tweet link
Software Development & Frameworks
@jezell · 2026-07-14T22:26
Custom Flutter embedder and custom Flutter engine on Skia Graphite with WebGPU bindings 🔥 https://t.co/ishj0qLsGw → tweet link
@jezell · 2026-07-14T20:11
Flutter Flame on WebGPU + WASM ✅ @spydon. Notes for the game say it might be slow because it pushes things. Nope. Buttery smooth in Skia Graphite. https://t.co/eSgIz66W9z → tweet link
@jezell · 2026-07-15T18:27
Web Embedder vs MacOS embedder. Same exact WASM app running inside the embedder on both platforms. WASM app contains custom Flutter + Skia Graphite engine. Native WebGPU is served by Dawn. https://t.co/VLrKjtPXv6 → tweet link
@jezell · 2026-07-15T16:58
WebGPU is all you need https://t.co/hkyH2gbNQH → tweet link
@jezell · 2026-07-15T16:19
RT @lancedb: Lance now opens a 65K column table in 17 milliseconds. It used to take 17s. That's 1,032x faster 🚀 Wide tables are no longer… → tweet link
@jezell · 2026-07-15T16:12
RT @alex_barashkov: Introducing Aval - a new open source format for interactive video on the web. It has a built-in state machine, frame ac… → tweet link
@steipete · 2026-07-15T04:34
Suno AI is delivering bangers! https://t.co/TUxkmyzsph → tweet link
@LinusEkenstam · 2026-07-15T15:08
RT @mukundjha: Most businesses don’t need just any software. They need a way to turn how they actually work into software. That’s what @eme… → tweet link
@levelsio · 2026-07-14T22:14
All you really need today to build a business is 100% free open source software that charges you $0/mo + a VPS server + an API to do some required AI stuff + some R2/S3 file hosting. They don't want me to tweet that because it destroys their businesses but I can't lie to you! → tweet link
@levelsio · 2026-07-15T16:02
I think you have to do something radically different as a landing page or app to even get anyone to use something these days as AI has made every landing page or app look the exact same → tweet link
Open Source & Local AI
@TheAhmadOsman · 2026-07-15T16:28
Proud to be featured in The Washington Post discussing why controlling model weights is becoming essential for organizations. At @OsmanticAI, we believe Opensource, self-hosted AI is not just an alternative but is the only architecture that makes longterm strategic sense. → tweet link
@TheAhmadOsman · 2026-07-14T23:15
Even Microsoft is hedging and talking about Sovereign AI and running the models weights on your hardware, and you still not bullish on Local and Opensource AI? → tweet link
@TheAhmadOsman · 2026-07-14T22:23
Local AI is easy now. - How easy??? Really easy... > curl -fsSL https://t.co/Y86A0lwP9P | bash. That's it, that's all you need now → tweet link
@alexocheema · 2026-07-15T18:35
If you don’t have at least 1TB VRAM and a VPN to torrent model weights from China, you are asleep at the wheel, and ignoring the uncomfortable truth that open weight models may be banned this year. Losing access to frontier models will be depressing... → tweet link
@ollama · 2026-07-15T04:21
Ollama is back in NYC today! It's exciting to see more people and businesses realize the benefits of open models! Ownership. Affordable. Private. Thank you @Nasdaq and the @Theoryvc team. → tweet link
@TheAhmadOsman · 2026-07-15T17:16
Local AI Summit? No. It's now called: > The Local AI Summer. More soon. https://t.co/qjYhQOMVMM → tweet link