Head of Insights @OpenRouter | Data Storyteller ex @cartainc, lover of data viz.
- Tweets 8,824
- Following 518
- Followers 21,729
- Likes 18,879
Pinned Tweet
State of Seed Startups - data from 10,000+ seed rounds from 2021-Q3 2025.
Valuations. Round Sizes. SAFEs vs priced. Dilution. Ai vs non-AI. Graduation Rates. SF vs everybody. Founder ownership. Hiring (+ AI/ML equity packages).
Full report: carta.com/learn/resources/st…
Data in 🧵
10
16
166
50,180
Soft launch - if you want insights from the 130 trillion tokens that flow through OpenRouter every week, join us.
Subscribe to Inferred, our data newsletter now.
openrouter.ai/data
4
4
22
7,268
Mary Meeker perfected the ultimate report style years ago. 200-250 charts, very little editorializing, smack me in the face with the numbers.
Who owns that mantle now?
22
4
102
16,793
Call out good charts when you see them - big fan of the bolding the 3-month average here but keeping the faded monthly spikes.
Gives the focus + the context. Nice one @ernietedeschi
from @stripe's eudata.vercel.app/
1
19
2,226
Peek into the first day for Opus 5.5 on OpenRouter:
- 22% of Anthropic tokens went to Opus 5.5 (best model launch by share in the last year)
- nearly double the user count as day 1 for Opus 5
- doing better today than yesterday
Claude cooked, will do a first week recap soon
5
2
228
10,370
Stage is set: Opus 5.5 and GPT-6 are live.
One view of the current landscape from us at OpenRouter below, comparing wallet share just among OAI and Anthropic models.
Will be fascinating to see how these new models change the game in the coming days.
2
6
94
8,870
Here's how Anthropic models were being used before today's Opus 5.5 launch.
Opus 5 took the most spend (~30% of all dollars that went to Ant models). It also earned 25% of tokens.
Let's see how fast Opus 5.5 wins over older model share
2
3
26
2,442
That new model smell 🍲
Claude Opus 5.5 from @AnthropicAI is live on OpenRouter!
The first model in the Claude 5.5 family leads Opus 5 and Fable 5.1 on agentic coding, knowledge work, and computer use, with 1M context at $4/M input and $20/M output, 20% lower per token than Opus 5.
ALT Benchmark table comparing Claude Opus 5.5 with Fable 5.1, Opus 5, GPT-6 Astra, and GPT-5.6 Sol on agentic coding, knowledge work, business workflows, multidisciplinary reasoning, scientific research, computer use, and chart recognition. Opus 5.5 leads on seven of nine rows.
2
1,609
When you talk about open-weight AI, you're really talking Chinese labs.
It's not 1:1 but it's pretty damn close (data below is share of tokens by week)
11
16
137
16,447
Swing and a miss
Dylan Patel (@dylan522p) of @SemiAnalysis_ says open source is dying:
"There's multiple Chinese model labs who are telling all the inference guys, 'Our next model's not going to be open source. We're going to license it to you.'"
" Open is dying quickly, unfortunately."
Community note
This July 2026 quote (recirculated without date) claimed Chinese labs were shifting away from open source. Multiple labs have since released large open-weight models, including Qwen updates in Aug-Sept 2026.
huggingface.co/blog/state-of-…
rits.shanghai.nyu.edu/news/
wired.com/story/chinas-o…
x.com/sourceryy/stat…
3
7
106
17,432
What should we make of all the data showing the growth of inexpensive open weight tokens?
8 thoughts, all personal takes:
1) This is not a death knell for closed frontier labs by any stretch. So and so lab is “screwed” is an easy way to farm engagement, not a considered opinion.
2) It does put to bed (for now) the idea of a single company or model dominating the AI market in perpetuity.
3) This market is in its infancy. Just this week, Jev was released and the cycle of innovation surged forward.
4) Different strategies will create different winners. Understandable that the focus is on trillion-plus valuations and whether those withstand scrutiny, but other players will pursue economic value in many different ways. Some may even avoid VC funding 😅
5) Hype cycles are only vaguely correlated to impact. X is where AI signal happens but it’s also intensely crowded with noise.
6) Inertia, especially in the short-term, is underrated. Most workloads don’t move every time a new model is released.
7) Hundreds of factors outside of cost and performance matter. One of these is “vibes” which are hard to quantify.
8) This may be the fastest growing pie in the history of capitalism. Tiny slices will still be incredibly lucrative.
These opinions will likely change as new data appears!
4
2
37
3,437
Look at the earnings revisions in 2026 and 2027. What changed?
Substantially different from any point in the past decade and it correlates basically exactly with agents becoming actually valuable.
6
13
101
14,216
There won’t be another day where open token volume fails to beat closed tokens.
The playing field has now shifted to spend.
(But also keep in mind this whole market is growing super quickly)
Looks like today may be a record day for token volume % of open models on Vercel AI Gateway:
🟦 Open 78.4% 🟨 Closed 21.6%
While spend 💲 usually tells a different story, #3 and #4 today are Moonshot AI & DeepSeek. Adding Z.ai, their combined spend surpasses OpenAI (#2).
(Do note that's the spend for inference of the model across providers (mostly in the US), not revenue going directly to the open weight labs.)
5
12
74
8,071
The argument over Anthropic in the FT today
a) Under pressure from OAI, open source, all fronts
b) the lynchpin of the AI bull case
Featuring @OpenRouter data of course :)
ft.com/content/96d0a206-a37b…
10
16
93
11,462
If for some reason you’re stuck on a waitlist…
Jev by @typesafeai is now on OpenRouter, in beta.
Jev is a System One model. Instead of generating text, it takes your app's state plus a typed question and returns a typed decision with a probability attached. There is no JSON prompting, parsing layer, and nothing to validate against.
11
3,255
