Chris
@ChrisGPT
Agi 2029 - AI Insider / Reporter as featured in Axios • The Information • NYT • Techcrunch
1.6K Following 68.4K Followers
There’s basically unanimous praise for Opus 5.5, but I think people *might* be overlooking what its existence implies about Anthropic’s internal models.
Anthropic almost certainly has substantially stronger internal models helping generate training environments. Think of the stronger internal model as the teacher and Opus 5.5 as the cheaper deployable student.
Now Opus 5.5 itself scores 55.8% on CoBench 2.1, while Anthropic estimates roughly 85% would be required to fully substitute for its research staff.
Opus 5.5 is now only - 30 percentage points away from Anthropic’s benchmark threshold for fully substituting its research staff.
Show more
Anthropic is sandbagging btw. Just like OpenAI both have models significantly more powerful than Opus 5.5 or Astra
Meta muse did launch a call feature but it’s outbound to US businesses only and half the time humans do it for you.
Mitra is ‘claiming’ it’s reliable and u can call anyone and it can answer too from someone’s own number while he whispered instructions to it live without interrupting the conversation.
This is a completely different abstraction compared to when Google had an AI agent answer a phone a couple years ago. (Never shipped to consumers btw)
These features shouldn’t be only for enterprise hopefully more Neo Labs can create pressure like this
Show more
Today we're launching Mitra.
Mitra is a personal AI assistant that lives on the phone. It makes and answers calls from your own number, handling the conversations you don't have time for. When it needs you, it asks. When a call needs you, it pulls you in. You can listen live, whisper instructions, or take over whenever you want. And when it's done, everything is waiting for you to review.
But calls are just where it starts. Mitra texts, emails, and messages on Slack as you, from your own accounts. It books appointments, keeps your docs and spreadsheets up to date, and handles tasks on the web using a secure vault for your logins.
The more you work together, the more it feels like you. It learns from what you share with it, the conversations it has, and the way you work. It remembers. It keeps things moving on its own and follows up before you think to ask.
Talking to it is simple. Text it, or just say what you need. You decide how it speaks for you and how much it handles alone. You can even build a small team of Mitras, each with its own personality and purpose, for every part of your life.
People are already using Mitra to stay close to the people they love, to start companies, and to run the ones they've built.
A personal AI only works if it's truly on your side, so we built our business around one thing: you. Mitra is a simple subscription. No ads, no selling your data, no one else it answers to. We only win when Mitra gets things done for you and represents you well.
Mitra is available today on iOS, iPad, and desktop. We'd love for you to try it.
Show more
Opus 5.5 & GPT 6 can cook together!
Within 18 months anyone in the world will be able to create their own Triple A game quality project.
And I’ve said in the past and I’ll say it again, I predict you will see statements / shareholder concern for the coming democratization of gaming.
You can tell the doomers and anti-AI people are getting worried when the sentiment shifts from "it looks like slop" or "it looks horrible" to "well, YOU didn't make this and the models made it." Now, even though you do need someone to steer it in a good direction, it is true - the models made this. Soon, the models will make AAA games better than your favorite studios, and people are really not going to take this well.
Show more
Opus 5.5, GPT-6 Astra, and Fable 5.1 recreate Zelda: Ocarina of Time !
We set out to recreate the first demo video of Zelda, and see what we could come up with in one week!
Everything you're seeing in this video was completely made from scratch without a game engine!
We could have pushed these models a lot further for a lot longer. However, we wanted to get this video out after about a week of working to show you guys how far the models have truly come!!
Show more
Opus 5.5, GPT-6 Astra, and Fable 5.1 recreate Zelda: Ocarina of Time !
We set out to recreate the first demo video of Zelda, and see what we could come up with in one week!
Everything you're seeing in this video was completely made from scratch without a game engine!
We could have pushed these models a lot further for a lot longer. However, we wanted to get this video out after about a week of working to show you guys how far the models have truly come!!
Show more
Anthropic is sandbagging btw. Just like OpenAI both have models significantly more powerful than Opus 5.5 or Astra
OpenAI will announce a ‘ProMax’ plan for 500$ per month on devday.
I’d be more than happy to buy this but if it doesn’t allow you to run any chats 24/7 (even zero subagents would be fine) I’d be quite disappointed
Show more
People are saying this is referring to hardware, - here’s what we know!
The prototype discussed in 2025 was described in reporting as screen-free, roughly pocket/palm-sized, and non-wearable.
And the More recent reporting in July 2026 says the first product is expected to be a portable, screenless smart-speaker/AI-companion type device with cameras and sensors.
Could this be it? Maybe. But I do not think they would tease hardware in this manner
This appears to me to be more of an Ultra Fast reference.
Show more
Opus 5.5, GPT-6 Astra, and Fable 5.1 recreate Zelda: Ocarina of Time !
We set out to recreate the first demo video of Zelda, and see what we could come up with in one week!
Everything you're seeing in this video was completely made from scratch without a game engine!
We could have pushed these models a lot further for a lot longer. However, we wanted to get this video out after about a week of working to show you guys how far the models have truly come!!
Show more
computer (Astra) lights on please.
this may be the most extensively detailed & accurate 3d scene ever created by an ai.
400+ hours of Astra
Today we’re launching micro1’s PII transformation model, flow-transform 1.0, delivering frontier-level performance across detection, identity synthesis, and transformation of personally identifiable information.
On PrivacyBench, our model reaches 96.0% F1, outperforming every detection baseline we tested, including Tonic Textual, Claude Opus 4.8, Sonnet 4.6, Microsoft Presidio, Haiku 4.5 and GLiNER2.
Some of the most valuable training data for frontier AI models lives inside fully functioning companies. It captures years of real work across decisions, communications, tools, handoffs, exceptions and the relationships connecting them.
The problem is that this data is also full of PII.
Traditional redaction makes the data safe, but it also destroys the very workflows and relationships frontier models need to learn from.
flow-transform 1.0 solves this by turning enterprise operational data into high-fidelity training data for frontier models by replacing real-world identities without flattening the reality the data captures.
Show more
Metas one more thing moment - they unveiled hardware for Meta Muse (beating OpenAI to an AI hardware device unveil)
Always on w/fingerprint sensor “by far the fastest way to talk to your muse” shipping in time for the holidays
I think OpenAI can one up this but we will see 👀
Show more
🚨 META JUST UNVEILED THE VR GLASSES
100 grams on your face - about the weight of a deck of cards. Meta says it has “the best display system that we have ever made,” and people in the demo were literally saying “it is like a computer” and “you could edit a whole movie just using the table.”
As Mark explained this is cinema, computer, and a game console all in one, with 3D space, virtual screens, hand occlusion, and DisplayPort over USB-C.
James Cameron’s reaction is probably the best quote in the whole segment “I’m inspired by what you’ve created. I’d like to see my work in these glasses.”
I am completely one shotted these look amazing!
Show more
For everyone wondering it’s GPT 6 Sol today - before Aeon.
The faster we get Opus 5.5 out of the way the faster we get to Fable 5.5
I built this browser-based drum machine / 16-step sequencer in about 15 minutes using Tencent’s Hy4 preview in WorkBuddy.
The model handled the frontend, sequencer logic, playback controls, BPM controls, and animated visualization from basically one build session.
Hy4 preview feels noticeably faster and more effective than Hy3, especially when iterating on frontend/code tasks like this.
@TencentHunyuan @TencentAI_News @WorkBuddy_AI
Show more
Wait? How the hell did Tencent compress Hy4 preview from 1.5TB down to 200GB and barely move the benchmarks??
They basically let calibration data decide how aggressively each layer can be compressed. Some layers got pushed all the way down to 1.31 bit, while the more sensitive ones stay closer to 2 bit so the model doesn’t fall apart. Literally insane efficiency gains.
And somehow MCP Atlas only moves down to 83.2, and SWE-Bench Multi - 82.9 to 81.3.
The low-bit inference progress happening right now is kind of insane.
Show more
🚨OpenAI Persistent Agent - Aeon is almost here!
I dug into Codex’s shipped code and found some more info about *Aeon*, OpenAI’s rumored Grok Bot competitor -
• Model-selection fields: `aeonModelId` and `allowAeonDraftModelSelection`
• Name and appearance fields: `aeonDraftName` and `aeonDraftAppearance`
• Cloud-environment selection
• A dedicated `thread/startAeon` operation
• Aeon references tied directly to the “Agents” sidebar
I’m assuming this is the groundwork for configurable agents inside Codex.
The community has reported Aeon references in ChatGPT’s web and Android code, as well - including `aeonId` and `memberAeonIds`, alongside purported `aeon-core` instructions describing a ‘persistent agent’ coordinating cloud tasks and automations.
GPT-Astra helped build it, and a launch could happen Thursday or DevDay!
The code references are verified in the build I inspected, some paths remain disabled.
Show more
Another day, another Robotics Lab delivering millions of hours of egocentric human experience and data.
Do you remember a couple years ago when everyone was complaining that we didn’t have enough data for robotics? Well, now it’s definitely getting to the point where that bottleneck is starting to disappear. This is why I’m glad there are so many NeoLabs popping up measuring body motion, object interaction, tactile force, 3D structure, etc. (and in Maxinsights’ case, capturing 3D structure instead of just raw first-person video).
Their scaling thesis is also pretty interesting an hour of someone repeatedly doing one clean task is not worth the same as an hour full of different objects, contacts, failures, recoveries, tool use, and corrections. So they basically think of it as effective experience = hours × information per hour, meaning the fuck ups and how humans recover from them are part of the valuable data too.
Show more
A little industry secret, every frontier lab has a tiny, tiny number of people who understand the architecture + training stack at a level almost nobody else does quite literally it can be as few as 1-6 people. Not just transformers on paper, but which changes actually survive trillion token training runs, how scaling behavior interacts with data mixtures, and all the tacit tricks that separate a good architecture from a frontier model, they can save a bad training bad and save the company millions and millions of dollars every training run.
Every lab has its “Noam Shazeers.”
When one of those people leaves, you’re losing years of accumulated, (largely undocumented) knowledge about how to make these systems actually scale and replacing that knowledge can materially set a lab back.
That’s why they’re are paid 100M to 1B in stock options + salary.
Show more
serious question, if Elon is so smart and his company so good, why cant he make a frontier model? I dont hate Elon or anything, im mostly neutral towards him, but he talks alot and his models are not good
Show more
Starting university today expecting to build a career around being the first human to crack major mathematical problems is starting to look increasingly precarious.
Terrance Tao - “a crisis in our mathematical values and practices.”
Other mathematicians are calling for us to slow down the math breakthroughs.
This is a very anti human way of looking at it. The breakthroughs coming will serve all of humanity. Don’t let ego get in the way.
Show more
We’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, and build tools that support mathematical research and learning.
Through this work, we want mathematicians to be at the center of shaping how AI supports mathematical understanding and how its benefits reach the wider community.
Show more
The Grok 4.7 & GPT 6 video are the same?
GROK 4.7 IS ACTUALLY COMPETING WITH GPT-6 ASTRA.
I gave GPT-6 Astra, Grok 4.7, Kimi K3 and Fable 5.1 the same prompt to build a flight simulator
Astra was still #1# overall, but Grok 4.7 was surprisingly close
Kimi K3 and Fable 5.1 were basically a draw and both produced a much smoother result, while Grok 4.7 was right up there with Astra in terms of overall quality.
overall: Astra > Grok ≈ Kimi ≈ Fable
GPT-6 Astra finally has some serious competition.
Show more
