About This Week’s Covers

This week’s cover is a nod to Anthropic’s frontier model Fable, and it is based on the classic illustration for Aesop’s Fables, drawn by Milo Winter, a spectacular illustrator in the early 20th century. In the original artwork, a child sits in a tree surrounded by animals. I changed the kid to the Figure 03 humanoid robot. The children’s cover is meant to be playful, as if the animals are entertaining the girl. But given the backdrop of AI, it has a foreboding feeling, with the wolf and the fox.

The category covers are a different theme because June 16th is Bloomsday, a commemoration of the Irish writer James Joyce. The novel Ulysses takes place over a single day, June 16th, 1904. Bloomsday is named after Leopold Bloom, the protagonist of Ulysses. I have a Claude skill that can take a simple theme and build category covers, and I gave it Bloomsday this week to see what it would do. My favorites are below:

This Week’s Humanities Selections

This week’s humanities reading commemorates Stephen Dobyns, an American poet who died on June 14th of 2026 at the age of 85.

“Each thing I do, I rush through so I can do something else. In such a way do the days pass—a blend of stock car racing and the never ending building of a gothic cathedral. Through the windows of my speeding car I see all that I love falling away: books unread, jokes untold, landscapes unvisited…”

“Hesitancy is the surest destroyer of talent. One cannot be timorous and reticent, one must be original and loud. New metaphors, new rhythms, new expressions of emotion can only spring from unhindered gall. Nothing should interfere with that intuition–not the fear of appearing stupid, nor of offending somebody, nor jeopardizing publication, nor being trivial. The intuition must be as unhindered as a karate chop.”

“A work of art gives testimony to what it is to be a human being. It bears witness, it extracts meaning. A work of art is also the clearest nonphysical way that emotions is communicated from one human being to another. The emotion isn’t referred to; it is re-created. The emotion shows us that our most private feelings are in fact shared feelings. And this offer us some relief from our existential isolation.”

“When the form is more important than the content, when it exists to convince or dazzle or decorate or distract, then the form is not rising out of the needs of the content but is being used rhetorically; that is, to convince for its own sake. The eventual effect will be to frustrate us, because we are looking, ultimately, to be moved out of ourselves, to be able for a second to step away and see ourselves in relation to the world.”

This week’s Humanities musical selection honors Abdullah Ibrahim, who died at the age of 91 this week. He’s a South African jazz pianist who goes all the way back to Duke Ellington. His friends included greats like Max Roach, Ornette Coleman (Dec 5, 2025 newsletter), Don Cherry, Cecil Taylor, Pharoah Sanders and Archie Shepp!

I chose the song “Brilliant Corners” because it’s avant-garde and distinctly human, especially since it’s from 1963. It’s also a cover of a Thelonious Monk song, which is a fun twist, since artificial intelligence is chided for being derivative.

Below is one of his performances as Dollar Brand. His birth name was Adolph Johannes Brand. He changed his name to Abdullah Ibrahim in 1968 after converting to Islam.

Thinking back to a younger version of ourselves seems fitting when reflecting on mortality. Ibrahim was incredibly innovative. I can only imagine what it would be like to play in his style in the ’60s. I wish I had a time machine to witness some of these people with such character and integrity of their art.

This Week By The Numbers

Total Organized Headlines: 521

This Week’s AI News Overview In Plain Terms (written by me, not AI, so it’s rambling!)

For the week ending June 19th, I hand-organized 521 links into about 55 categories. I’m still 12 weeks behind because I enjoyed the summer with my family.

Moments of Touching Grass
I continued training for my attempt to climb Grand Traverse Peak with my daughter in August. I ran in the soft sand a few times and swam in the ocean to cool down, and I continued crazy StairMaster workouts.

On June 17th, I climbed 300 flights with a 15-pound pack without using my hands. That was up 10% from the previous week, and it’s about 3,214 feet. It’s quite a long time on the machine, but it’s fun to have a goal. Grand Traverse is a 13,063-foot mountain, and the hike starts at 8,120, just under 5,000 feet in gain.

This Week’s AI News Summaries!
I’ve organized the AI news this week mostly by company, lightly sorted by priority. The big story this week continues to be Anthropic’s new models, Mythos and Fable.

Mythos is Anthropic’s truly frontier model, and it is so powerful that they will not release it to the public.

Last week, Anthropic attempted to release a nerfed version of Mythos called Fable (Mythos with security guardrails). Anthropic is getting pulled in two when it comes to the PR game… because the public doesn’t understand what’s happening, and no one trusts anyone they’re hearing from the tech folks. Consumers are either cynical and think it’s marketing BS, or they’re terrified and think everything needs to be paused. Little middle ground.

In the meantime, the Trump administration is still upset about Anthropic’s refusal to allow the government to use AI for surveilling citizens and/or autonomous weaponry.

Over the past few weeks however, several government agencies have been looking at Mythos as part of a secure study called Project Glasswing. Public and private experts have now concurred with Anthropic’s assessment that Mythos is too powerful to let loose.

Separate from the supply chain risk declaration by Hegseth a few weeks ago, the US government this week mandated that no foreign national anywhere can use Fable or Mythos. That essentially shuts down all enterprise use because large companies have subsidiaries across the world. It also ties directly into the timing of the G7 summit this week… because not even G7 partner countries can use Anthropic’s most advanced models.

Developers of middleware like Cognition have had to remove Fable from their products, and it’s a little chaotic while everyone tries to figure out what’s going on.

Statement from Anthropic:
https://www.anthropic.com/news/fable-mythos-access

MidJourney Medical Pivot
The second big story this week is that image generation company MidJourney has pivoted into a full-body scanning medical imagery startup.

MidJourney’s founder, David Holz, is a bit of a legend. He worked at the Max Planck Institute and also NASA’s Langley Research Center. He did graduate research in neuroscience and atmospheric science, and he has a degree in applied math from UNC Chapel Hill.

The new MidJourney scanner is a giant vat of liquid, you descend into it, and it maps your body, using some sort of wavelength analysis. It looks like a CAT scan or an MRI. The medical scan gallery for MidJourney uses CAT scans and MRIs as counterpoints to the resolution.

MidJourney Spas: https://www.midjourney.com/medical/spa_gallery

SpaceX Buys Cursor
SpaceX bought the software development platform Cursor for $60 billion bucks. What an amazing ride for Cursor. Cursor was founded four years ago with an $8 million seed round. As of May 2026, their revenue was $3 billion, and SpaceX just bought them for $60 billion…

Research and Benchmarks
Pew Research came out with a survey on Americans’ usage and views on chatbots and AI. About half of Americans are now using chatbots, but overall sentiment is still very negative. The research has a lot of other data points, like 40% of U.S. adults now have a smartwatch and about one-third have a smart speaker. It’s worth skimming to see the state of things.

Epoch AI came out with statistics showing that the maximum computing capacity for any single data center is doubling every seven months.

Another Epoch chart shows that hyperscalers may be running out of cash. A hyperscaler is one of a handful of enormous cloud-based systems that allow users to rent data storage quickly…. Microsoft, Amazon, Alphabet, Meta, Oracle.

Small Models Crushing Souls
ChatGPT’s Instant model can match advanced thinking models when it comes to health questions.

This is an important trend to follow. For example, open-source software is always about six to twelve months behind the frontier. The price of AI quality goes down quickly over time. So, if a company is spending a ton of money to be the absolute best at AI… six months later there’s a cheap alternative to their product, which makes it a very difficult pace to maintain.

Even within the frontier labs, there are competing small models chipping away at the top model (by design). Just like cars have trim levels.

For example, Anthropic uses a naming convention where their smallest model is called Haiku (just like the poem is a small poem). The next strongest/sized model is called Sonnet. The next largest is called Opus… then Fable…and then the absolute best frontier model is Mythos.

Anthropic times its releases in an alternating cycle. Sonnet is better than Haiku, but the next version of Haiku is as good as the previous version of Sonnet.

This is what’s happening when you see a headline like ChatGPT Instant is able to match a thinking model. Instant means that it’s a very fast response, and needs less thinking and less work. The longer a model thinks, the more computing power and tokens it’s using. So if a model can quickly respond, it’s gonna be a lot cheaper.

Every week 230 million people use ChatGPT for health and wellness questions.

And… ChatGPT Instant is getting good. So even if you don’t pay for ChatGPT, your experience is going to be getting strong.

Following the pattern… the people using the paid plans now are perhaps about a year ahead of the people who are using the free plans, but the free plan gets better over time.

One of my favorite benchmarking companies, Epoch AI, corrected some of their frontier math benchmarks this week. OpenAI GPT 5.5 Extra High doubled after the adjustment. It’s one of the rare times you see a benchmark performance go up after review instead of down.

ChatGPT Real-Time 2 has been out since May, and just like “instant” has improved… real-time voice has gotten faster and more powerful. If you’re used to using ChatGPT’s real-time voice, you’ll notice that it buys itself time by saying “ah-ha” or “um.”

That said, it’s fast. And if you use it as an interface, ChatGPT Real-Time 2 allows you to use your voice to control your desktop apps.

Marketshare
TechCrunch reports that ChatGPT’s market share is below 50% for the first time. Second place is Google Gemini. Amazingly, Claude is far below in third place.

But even if ChatGPT is below 50%, it’s significantly ahead of Gemini. GPT is at 46%, Gemini is at 27%, Claude’s only at 10%. That means that Grok, Perplexity, DeepSeek, and Meta all have less than 5% share.

Big hire for OpenAI
The T in ChatGPT stands for Transformer. Noam Shazeer is one of the co-inventors of the Transformer, and he has left Google DeepMind to go work at OpenAI.

Financials
Leaked data shows that OpenAI might be profitable serving customers with a 40% gross margin. However, that does not include training.

More OpenAI
OpenAI worked with Boston Children’s Hospital and Harvard and published a study that showed how O3 Deep Research was able to help doctors gain insights around rare pediatric diseases.

It wouldn’t be a week in AI news without a few blog posts from frontier labs. OpenAI posted “Reinforcement Learning Towards Broadly and Persistently Beneficial Models.” The undertones of this are always terrifying.

Google researchers released a paper outlining four paths from artificial general intelligence to superintelligence. It blows my mind (in a bad way) because they didn’t say anything new, but they published it as an arXiv paper.

I had GPT make a cartoon of it if you want to read a summary what Google wrote. It’s a valuable cartoon, even though it’s busy and looks like slop. All the elements are necessary, and it explains the paper very well. I purposefully made it a bit like Magic School Bus… that was me.

Nothing in the cartoon is necessarily new or paper-worthy. I’m not sure why it merited a paper, other than Google wanted to get on the record about what they think is going to happen.

At the G7 summit this week, the big frontier labs, including Anthropic and Google DeepMind, suggested a U.S.-led coalition of frontier development, aligning themselves as allies in the race, or the space race, for AI dominance.

Alibaba announced a new family of foundation models designed to help create embodied robots. Qwen is Alibaba’s frontier AI model, and it’s open-source. If last year was all about agents, next year is probably going to be about robots.

Also in open-source news, DeepSeek raised $7.4 billion and is now China’s most valuable AI startup.

More open source this week: ZAI released GLM 5.2, which is now the number one ranked open-source model in the world.

Microsoft is considering adding DeepSeek to its Copilot suite as a cost-saving option.

McDonald’s announced that they’re using a Google-powered AI ordering system in a handful of franchise locations as a test. I wish I knew the locations because I would love to try one.

And finally, Microsoft launched Copilot Co-work. It’s basically the same as Claude Co-work or OpenAI work mode…. and provides continuity across a project and files. It also includes plugins, just like connectors in GPT and Claude.

This Week’s Top Stories With All of the Links

Anthropic

Mythos Safeguards: Anthropic is nerfing Mythos and people don’t get why or trust them at all. We have entered the chaotic phase.

Two things are true: (1) Anthropic (or parts of it) are absolutely and sincerely worried about the misuse of Mythos-class models & have put in excessive safeguards until they are confident it will not be misused (2) They have not succeeded in explaining/convincing people of this
https://x.com/emollick/status/2065115360148197630

White House Update: Trump administration won’t let any other country use Anthropic’s best models, Mythos and Fable. Not even allies.

Anthropic says Mythos and Fable 5 access could return in coming days after White House block
https://www.koreajoongangdaily.com/business/anthropic-confident-of-reenabling-mythos-fable-5-access-in-coming-days-executive/12727522

Anthropic, the U.S. government, and the Mythos/Fable shutdown – a roundup of the latest news: – Senior Anthropic staff are reportedly meeting with government officials in Washington today to try and diffuse the situation – Axios sources framing the issue as a communication
https://x.com/TheRundownAI/status/2066559132963131523

Axios reports that the industry is now worried White House export controls on Anthropic’s latest model could hurt the entire U.S. AI industry. The problem is trust. And that was to be expected. As Deutsche Bank’s Jim Reid put it: “You can’t rely on something that could be
https://x.com/kimmonismus/status/2066882221198245939

Exclusive | Trump officials won’t allow G7 countries to access Anthropic’s most advanced AI models: ‘Completely illogical’
https://nypost.com/2026/06/16/business/trump-admin-open-to-talks-with-anthropic-over-foreigner-ban/

Interesting: The US Commerce Secretary reportedly told Anthropic it needs government permission to export Fable 5 and Mythos 5 anywhere in the world. Even to “any foreign national regardless of location.” “”Lutnick threatened Anthropic with criminal and civil penalties if it
https://x.com/kimmonismus/status/2066972690926522593

Statement on the US government directive to suspend access to Fable 5 and Mythos 5 \ Anthropic
https://www.anthropic.com/news/fable-mythos-access

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of
https://x.com/AnthropicAI/status/2065597531644743999

UK Prime Minister Keir Starmer reportedly asked the Trump administration for a carve-out so British nationals and companies could regain access to Anthropic’s most advanced AI models. The answer was no. A Trump administration official told The NY-Post that giving even a G7 ally
https://x.com/kimmonismus/status/2066934409840775201

Update: We’ve removed Claude Fable 5 from Arena, following Anthropic’s latest announcement and the U.S. government directive to suspend access. Claude Fable 5 is the most powerful model we’ve ever tested – ranking #1 across Agent, Text, and Code Arena, and setting a new
https://x.com/arena/status/2065620808773611997

We’re removing access to Claude Fable 5 in Cognition products following Anthropic’s latest announcement and the US government directive to suspend access. All other models remain available in Devin, including Claude Opus 4.8 and GPT-5.5. Devin Ultra mode will continue to use
https://x.com/cognition/status/2065609115939062197

Gemma Optimization: Fable 5 maxes Gemma 4 at 255 tokens per second in browsers

Before Fable 5 was shut down, it pushed Gemma 4 to 255 tok/s on WebGPU. Some didn’t believe it was real. Today we’re releasing the demo and kernels it wrote for you to see yourself. Run it locally in your browser. Agentic kernel optimization is the future of on-device inference
https://x.com/xenovacom/status/2067289897111638484

How Claude Code is Used: A guide on how the pros use Claude Code

How Claude Code is used in practice \ Anthropic
https://www.anthropic.com/research/claude-code-expertise

Improvement on Math: Claude Fable 5 hits 87-88% on FrontierMath benchmarks

The shape of the graph is getting very familiar. Claude Fable 5 scores very well on FrontierMath: Tiers 1–4 (v2), reaching 87% on Tiers 1–3 and 88% on Tier 4. This continues a streak of Anthropic models improving rapidly at math.
https://x.com/emollick/status/2065513498763800850

MidJourney

MidJourney Medical Scanner: Image creation company Midjourney goes HAM with a full-body ultrasound scanner and medical imaging tools

A New Era of Midjourney
https://www.midjourney.com/medical/blogpost

A technical dive inside our new “”Midjourney Scanner””
https://x.com/midjourney/status/2067422898407837797

Gallery — Midjourney Medical
https://www.midjourney.com/medical/scan_gallery

I put my hand in the @midjourney demo scanner tonight and I felt the blessing of the future
https://x.com/saranormous/status/2067434920134549524

Midjourney Medical | Hacker News
https://news.ycombinator.com/item?id=48579650#48580452

Midjourney, The AI Image Generator, Is Developing A Full-Body Ultrasonic Scanner
https://www.engadget.com/2196998/midjourney-full-body-ultrasonic-scanner/

The Spa — Midjourney Medical
https://www.midjourney.com/medical/spa_gallery

this is so amazing. let inventors like @DavidSHolz invent
https://x.com/saranormous/status/2067435669694374217

SpaceX

Curson Acquisition: Big story: SpaceX acquires AI coding tool Cursor for $60B

Official now: SpaceX is officially buying Cursor in an all-stock deal valued at $60B. The move gives Elon Musk’s AI empire a serious enterprise coding product overnight, and could help xAI close the gap with Anthropic’s Claude Code and OpenAI’s Codex. However, the truth is that
https://x.com/kimmonismus/status/2066863066898116954

SpaceX acquired Cursor for $60B. A lot of people dismissed Cursor as just a “GPT/Claude wrapper” and a VS Code fork. But it became a company capable of post-training its own models. That is a big reason Elon wanted it. I think more companies will post-train custom models to
https://x.com/Yuchenj_UW/status/2066891492187320405

SpaceX has exercised the option to acquire @cursor_ai in an all-stock transaction with the goal of building the world’s most useful AI models. For the past few months, SpaceXAI has been jointly training a model with Cursor, which will be released in Cursor and Grok Build soon.
https://x.com/SpaceX/status/2066873915717136548

Public Opinion

49% of US using chat bots: Half of US adults use AI chat bots. Negative perception continues to rise.

Americans’ Views on AI Chatbots, Smart Devices and AI’s Impact | Pew Research Center
https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/

Data Centers

Computing Capacity Doubling Every 7 Months: Data center computing capacity is doubling every seven months

The record for computing capacity in a single data center has doubled every 7 months. Colossus 1, Anthropic-Amazon New Carlisle, and Meta Prometheus have each claimed the top spot in turn.
https://x.com/EpochAIResearch/status/2065131037324402744

End of Funding?

AI Capex Running Out of Money?: Hyperscalers on track to outspend cash flow on AI by the end of the year (virtual data hosts like Amazon Web Services, Microsoft Azure, Google Cloud Platform, Alibaba Cloud)

The end of the self-funded AI buildout? Hyperscaler cash capex is growing much faster than cash inflows. On current trends, they will be unable to fully fund the AI infrastructure buildout with cash from operations by the end of this year.
https://x.com/EpochAIResearch/status/2066955223437058115

OpenAI

GPT-5.5 Instant: ChatGPT’s Instant model now matches “thinking” models on health questions

GPT-5.5 Instant is now on par with our frontier Thinking models for health-related questions. Every week, more than 230 million people turn to ChatGPT with health and wellness questions, and GPT-5.5 Instant is better at recognizing when urgent care may be needed, asking for
https://x.com/OpenAI/status/2067672740539306261

GPT-5.5-xhigh: EpochAI benchmark correction essentually doubles GPT-5.5-xhigh’s FrontierMath score

GPT-5.5-xhigh’s FrontierMath 4 score jumped from 35% to 73% after EpochAI fixed errors in the benchmark
https://x.com/scaling01/status/2065490265691902415

GPT-Bidi-1: OpenAI is launching GPT-Bidi-1 for yet another improvement to ChatGPT voice (which is getting pretty strong)

OpenAI prepares major ChatGPT voice upgrade with GPT-Bidi-1
https://www.testingcatalog.com/openai-prepares-major-chatgpt-voice-upgrade-with-gpt-bidi-1/

GPT-Realtime 2: Speaking of voice…. GPT-Realtime 2 enables full voice control of desktop apps! Let’s go.

GPT-Realtime 2 is the future of the operating system. I’ve been experimenting with it for a couple weeks now, and I gotta say, it’s pretty gosh darn incredible. Opening apps, searching the web, even editing in Premiere. All with just my voice. And it only takes a few prompts
https://x.com/per_simmons_/status/2067051453022363992

Market Share Slips: ChatGPT market share drops below 50%

ChatGPT’s market share slips below 50% for first time | TechCrunch
https://techcrunch.com/2026/06/16/chatgpts-market-share-slips-below-50-for-first-time/

Noam Shazeer: Transformer co-inventor Noam Shazeer leaves Google DeepMind for OpenAI

I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there. It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to
https://x.com/NoamShazeer/status/2067400851438932297

Noam Shazeer – co-author of the Transformer, T5, and Switch Transformer papers, and one of the pioneers of sparse MoE models – is leaving his VP Engineering / Gemini co-lead role at Google DeepMind to join OpenAI This is likely the most significant AI talent move of the year.
https://x.com/scaling01/status/2067407207956705517

OpenAI Financials: OpenAI reportedly running at 40%+ gross margins

If the leaked financial data is right, OpenAI is profitable on serving customers with 40%+ gross margins. But training remains incredibly expensive. Automating AI research may also be a play to increase the efficiency of training: a superhuman researcher could do more with less.
https://x.com/emollick/status/2067258683436798073

Pediatric Disease Breakthroughts: OpenAI’s o3 Deep Research helps solve rare pediatric disease cases

Together with researchers at Boston Children’s Hospital and Harvard, we published a study in NEJM AI showing how o3 Deep Research helped clinicians revisit previously unsolved rare pediatric disease cases, and find answers for families who had waited years.
https://x.com/OpenAI/status/2067625110199247353

Reinforcement Learning: Reinforcement learning towards broadly and persistently beneficial models… gotta love these titles

Reinforcement learning towards broadly and persistently beneficial models
https://alignment.openai.com/beneficial-rl/

Valuation: OpenAI surpasses Anthropic in latest private market valuation (a reference to Noam leaving Google)

Breaking: OpenAI overtakes Anthropic’s valuation
https://x.com/LiamFedus/status/2067415148697116980

Google

What Happens After AGI: Google DeepMind maps four paths from AGI to superintelligence, but you have to read a arxiv paper… so I’ll make a cartoon

.@GoogleDeepMind just released a 57-page report on a question that not many people discuss now: What happens after AGI? → ASI (artificial superintelligence) systems more capable than large groups of human experts DeepMind sees 4 possible paths from AGI to ASI: • Scaling
https://x.com/TheTuringPost/status/2066321270220918870

Government

G7: Anthropic Google DeepMind CEOs call for U.S.-led AI coalition at G7

AI in spotlight at G7 as Trump, world leaders joined by tech chiefs
https://www.cnbc.com/2026/06/17/g7-trump-ai-tech-leaders-openai-anthropic-google.html

Anthropic Google DeepMind CEOs call for U.S.-led AI coalition at G7
https://www.cnbc.com/2026/06/17/anthropic-amodei-google-hassabis-us-ai-coalition-g7.html

Qwen

Robots: Alibaba announces Qwen-Robot family of embodied AI foundation models. 2026 was agents. 2027 is robots.

[2606.17030] Qwen-RobotWorld Technical Report: Unifying Embodied World Modeling through Language-Conditioned Video Generation
https://arxiv.org/abs/2606.17030

📣 Introducing the Qwen-Robot Suite — Qwen-RobotNav, Qwen-RobotManip, Qwen-RobotWorld, three foundation models, a full stack for embodied intelligence. 🧭 Qwen-RobotNav — the gateway to mobility. • Unifies 5 navigation tasks in one model: instruction following, point-goal,
https://x.com/Alibaba_Qwen/status/2066870197122899980

Zai

$7.4 Billion Fundraising: DeepSeek raises $7.4 billion, becomes China’s most valuable AI startup

DeepSeek Becomes China’s Most Valuable AI Startup After $7.4 Billion Fundraise – WSJ
https://www.wsj.com/tech/ai/deepseek-becomes-chinas-most-valuable-ai-startup-after-over-7-4-billion-fundraise-78ef64c0

GLM 5.2: GLM-5.2 is the newest #1 open-source model. It might not beat frontier models, but it’s crazy how good open source is getting

@jeremyphoward @Zai_org GLM-5.2 is between GPT-5.5 and Opus 4.8 in our new agentic knowledge work eval that we released just today. Very impressive
https://x.com/ArtificialAnlys/status/2067761754990686483

BREAKING: GLM-5.2 is now 1st on Design Arena. With an Elo of 1360, GLM-5.2 has jumped ahead of the now unavailable Claude Fable 5. And it’s open weights. This is an improvement of 4 positions and 27 Elo points to achieve one of the highest Elo scores in our code categories
https://x.com/Designarena/status/2066940737011560652

Credit to GLM-5.2 Max, the new open weights model, for pulling this off. …but you can see the difference between it and Fable in a way benchmarks don’t show. GLM-5.2 gives a correct poem (& the Welsh is fun) but Fable weaves the disappearing letters into the theme of the poem.
https://x.com/emollick/status/2067056226337186146

Exciting news: GLM-5.2 (Max) ranks #2 in Code Arena: Frontend, with +29pt over Claude Opus 4.7 (Thinking) and only behind Fable 5! GLM-5.2 is the best open model vs Kimi-K2.6 and Minimax-M3 by a large margin. – #2 React and #4 HTML sub-leaderboards – Ranks as the top model in
https://x.com/arena/status/2066957802741043641

GLM 5.2 ranks #3 on FrontierSWE. It is only behind Fable 5 and Opus 4.8, and it outperforms GPT-5.5. This is the first model that closes the large gap between models from Anthropic / OpenAI and other providers, and it is the strongest open-weight model by far.
https://x.com/ProximalHQ/status/2066939701026787583

GLM-5.2 (Max) by @Zai_org ranks #10 on the new Agent Arena leaderboard, closely matching Claude-Opus-4.8 (non-thinking) and is the #1 open model by a wide margin! In Agent Arena, we measure models on millions of real-world, long-horizon agentic tasks from a global community of
https://x.com/arena/status/2066943450914943025

GLM-5.2 is the first open-weights model to cross 80% on Terminal-Bench, and beats every other open model available. It also beats Gemini, making it a frontier-level model for a fraction of the cost. Open weights is back. This model is a game changer. Available in Cline now!
https://x.com/cline/status/2066951439793242193

GLM-5.2: Built for Long-Horizon Tasks
https://z.ai/blog/glm-5.2

Introducing GLM-5.2: Frontier Intelligence, Open Weights – Significant improvements in coding and agentic tasks – Strong long-horizon capabilities with a 1M context window – Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong
https://x.com/Zai_org/status/2066938937344495629

It’s hard to pinpoint open-closed gap and so-on, but I trust the @arena team and just look where GLM 5.2 is on this. An MIT licensed, to be open weight model. At this point you could argue they have a better agent than Gemini does. That’s a serious accomplishment.
https://x.com/natolambert/status/2066968753221624303

Microsoft

DeepSeek for CoPilot: Microsoft is thinking about including DeepSeek V4 to make Copilot cheaper

Axios reports that Microsoft is considering a Microsoft-hosted version of DeepSeek V4 as a cheaper model option for Copilot Cowork. Microsoft says Copilot Cowork can’t work on unlimited pricing. “We have users who do hundreds of tasks a week… but the consequence is the costs
https://x.com/kimmonismus/status/2066946013026263110