Sep 30, 2026 · 17 min · 8 segments
Watch the video version: https://www.youtube.com/watch?v=OZ7\_LtMCqocGPT-6.1 Sol vs Sonnet 5.5: $2 AI Price War + Anthropic's $2T IPO SecretsThe AI price war just went nuclear: OpenAI's GPT-6.1 Sol…
Well, we have to step back and look at the sheer scale of the technological triumph we just witnessed this week.
The price-to-performance ratio achieved by GPT 6.1 Sol and Sonnet 5.5 is... it's an objective victory for democratized innovation.
If you just look at the benchmarks, sure.
But consider those benchmarks.
Sol has essentially tied the flagship GBT six Astra model on the DeepSuite version one point one coding benchmark, scoring, uh, seventy-five percent against Astra's seventy-four point eight.
Mm-hmm.
But I wanna focus on the OS World 2.0 benchmark for a second.
Right, because that's the one testing agent to computer tasks.
Exactly.
For those following along, OS World doesn't just test if an AI can answer a text question.
It tests if the AI can take over a virtual computer's mouse and keyboard to, you know, open spreadsheets, navigate web browsers, and execute complex enterprise tasks completely on its own.
Which is a massive leap in capability.
It is.
And Sol scored within two point one points of Astra on this.
But here's the kicker.
It completed those tasks for roughly one dollar and thirty cents compared to Astra's nine dollars and thirty cents.
We are seeing a ninety-five percent discount on what they call cached inputs.
It brings the cost down to just ten cents per million tokens.
Which, just to ground that mechanically for a second, means the AI doesn't have to re-read the entire conversation history every single time you ask a follow-up question.
The system temporarily memorizes the context.
Right, it caches it.
Yeah.
It's a massive computational shortcut that slashes the processing cost.
Precisely.
I mean, the mechanical efficiency is staggering.
Well, we have to step back and look at the sheer scale of the technological triumph we just witnessed this week.
The price-to-performance ratio achieved by GPT 6.1 Sol and Sonnet 5.5 is... it's an objective victory for democratized innovation.
If you just look at the benchmarks, sure.
But consider those benchmarks.
Sol has essentially tied the flagship GBT six Astra model on the DeepSuite version one point one coding benchmark, scoring, uh, seventy-five percent against Astra's seventy-four point eight.
Mm-hmm.
But I wanna focus on the OS World 2.0 benchmark for a second.
Right, because that's the one testing agent to computer tasks.
Exactly.
For those following along, OS World doesn't just test if an AI can answer a text question.
It tests if the AI can take over a virtual computer's mouse and keyboard to, you know, open spreadsheets, navigate web browsers, and execute complex enterprise tasks completely on its own.
Which is a massive leap in capability.
It is.
And Sol scored within two point one points of Astra on this.
But here's the kicker.
It completed those tasks for roughly one dollar and thirty cents compared to Astra's nine dollars and thirty cents.
We are seeing a ninety-five percent discount on what they call cached inputs.
It brings the cost down to just ten cents per million tokens.
Which, just to ground that mechanically for a second, means the AI doesn't have to re-read the entire conversation history every single time you ask a follow-up question.
The system temporarily memorizes the context.
Right, it caches it.
Yeah.
It's a massive computational shortcut that slashes the processing cost.
Precisely.
I mean, the mechanical efficiency is staggering.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.