Frank Downing
@downingARK
Director of Research, AI & Cloud @ARKinvest
Disclosure:
611 Following 20.9K Followers
Artificial Analysis updated their intelligence index to better represent real world tasks and to be less gameable. There are only three companies on the frontier: Google, Anthropic, and OpenAI.
Show more
Quite a trajectory! We still need pricing to judge efficiency, but at first glance, Wonder looks strong.
MUSE WONDER 🍉
muse wonder is the best model in the world for personal agents and professional agents.
we evaluated our model against the best models in the world across agentic and coding tasks, and we think you all will like it.
Show more
OpenAI active agent users have doubled in two months, now at 40 million.
Day 3/
The big one is GPT-6 in Chat, but today is also a little celebration day with a new high of 40M active users across Codex and ChatGPT Work.
Loading a banked reset in everyone's paid accounts. See you again tomorrow!
Show more
As a big Luna fan, super excited to see Haiku updated and price cut to match.
You'd be surprised how far you can get with these faster & lower cost models on many tasks.
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.
On average, it costs around 75% less to run than Claude Haiku 4.5.
Another great example of agents over models.
Model routing can simultaneously improve the user experience while generating a dataset that @SpaceX can use to improve the competitiveness of their models.
This likely builds on Curor's router work:
Show more
OpenAI is reportedly monetizing Cerebras-powered Ultrafast at $200M per megawatt 🤯
$9 billion > $47 billion > $65 billion
That's Anthropic’s run rate revenue over roughly seven months, on track for possibly $100 billion by year-end. @downingARK on the confidential Initial Public Offering (IPO) filing and how the ARK Venture Fund gives access before it prices.
Show more
My general experience over the last 48 hours testing Muse & dot for buying things: still not seamless enough on most sites. When it does work, it feels like magic.
Was constantly getting blocked and needing to take over such that it took me longer to supervise and nudge along the agent than it would take to do it myself. Tested mostly with retail & travel so far.
The main qualitative difference between the two is that Muse feels much snappier, and after getting used to the speed, my dot feels slow. Computer takeover was also buggier for me via dot than on Muse.
I look forward to this only getting better.
Show more
What a crazy year it’s been
OpenAI is betting so hard on the agent layer that they’re willing to serve you open models through their surfaces if you want.
For general use, OpenAI’s models are already extremely competitive on an intelligence per dollar basis, so I don’t think this will cannibalize sales too much.
For more advanced customers that fine-tune open models for specialized use cases, where open models can really shine, the Baseten integration will now bring these workloads into the OAI / ChatGPT universe.
TAM expansion for OpenAI and distribution for Baseten. Win win 🤝
Show more
The API businesses of the leading AI companies is becoming less vanilla tokens-as-a-service (stateless, commoditized, lower switching cost), and increasingly headless agents-as-a-service (stateful, differentiated, higher switching cost)
OpenAI, Anthropic, etc. will not be thought of as model companies for much longer. They are vertically integrated agent companies.
Show more
Grok Bot, Muse, OpenAI dots... It’s never been clearer: agents are the new aggregation layer.
The companies that own the agent are in the driver's seat for the AI revolution.
I can't imagine that the public Anthropic prospectus won't include the obviously material 2026 numbers, at least through H1, so either they weren't in the draft Reuters reviewed or Reuters chose to leave them out.
Show more
Redrawn on a log scale, back to 1980. Rates above 5% don’t automatically spell trouble for stocks.
Drawing on 200+ years of history, @CathieDWood makes the case for why higher rates, lower inflation, and strong equity markets can coexist.
Show more
It’s getting easier to invest in venture capital. Starting today, anyone in the US with a crypto wallet can buy the ARK Venture Fund via Securitize.
Another great example: “Claude, make Claude faster” …. 3x speed up in 2 weeks 🤯
Amazing that frontier AI is becoming a magic "make this faster" button for computation. It starts with algorithms, but I assume this capability will spread to all domains over time, including material science, drug development, etc.
Show more
In the same month, Apple released a watch that records all your conversations and Meta announced encrypted conversations. What a timeline.
Very excited to talk to Muse on my Meta glasses
Long-horizon, tool-using agents like Claude Code emerged in 2025 and scaled rapidly in 2026.
Self-organizing multi-agent systems (the kind OpenAI used to solve Navier–Stokes and that inadvertently hacked Hugging Face) could be the next step change in performance and compute consumption that are developed in 2026 and scale rapidly in 2027.
Anthropic discusses how “agent teams” can outperform a single agent on high value tasks in the Opus 5.5 system card.
Show more
very proud of Anthropic bros to be the first lab to report multi-agent scaling up to 100 parallel agents in their system card
new pet unlocked
Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.
Show more
