The trick: Moved Ruler
Asana says keeping its browser agent's cache intact cut costs 76x.
The 76x also counts a switch from Model B to GPT-6.1 Sol.
Asana says keeping a browser agent's cache intact cut its cost 76x and made it 5x faster. A Reddit post retelling it adds that the change was not a new model.
Before you read on. Your call?
TRUE, BUT
76
OpenAI's page gives the basis: the optimized workflow on GPT-6.1 Sol averaged $0.47 in estimated model costs per run, 76x cheaper than the original production setup on Model B.
The twist
Within one model, the cuts are smaller. On Model B, the best condition cut cost per run 29x, per Asana. On GPT-6.1 Sol alone, at the larger history budget, the caching and screenshot policy cut cost 4x, from $1.97 to $0.47, per OpenAI's page. Asana's own post says that without batch pruning, caching the history at the larger budget cost more than not caching it on three of four models.
There’s more to this story.
Membership opens the full investigation, the strongest counterargument and what to do with what you’ve learned.
30 days free, then A$89 a year. Payment card required. Cancel before the trial ends to avoid the first charge.
Review my 30-day free trial →- At sign-up: 30 days free. Payment card required. One introductory trial per customer.
- Your trial ends 30 days after you start. Unless you cancel before then in Account → Manage subscription, we charge A$89 for the first year. It renews automatically at A$89/yr until cancelled.
- You can ask for a full refund within 14 days after any annual payment, renewals included, with no reason needed, by emailing hello@bskiller.com. This voluntary refund does not limit your rights under the Australian Consumer Law.
By starting your trial, you agree to the Terms.
BS Killer is published by Inferno Tech Pty Ltd, ABN 27 647 413 474.
Get tomorrow’s check free by email.
Couldn't check your access. That's on us.
The trick has a name
We call it Moved Ruler: two methods, two answers, one of them quoted. You'll see it again. Learn to spot it →
Receipts
- Supports asana.com:
How we cut a browser agent's cost 76x and made it 5x faster by keeping its cache intact
- Supports web.archive.org:
The change was not a new model.
- Supports asana.com:
On GPT-6.1 Sol, the same agent cost 76x less and ran 5x faster, reading 89% of its input from cache.
- Supports web.archive.org:
Asana cuts model costs 76x in browser tests with GPT‑6.1 Sol
- Refutes web.archive.org:
averaged $0.47 in estimated model costs and about four minutes per run, 76x cheaper and 5x faster than the original production setup on Model B.
- Refutes web.archive.org:
On GPT‑6.1 Sol alone, with the larger history budget, the new caching and screenshot policy reduced cost 4x, from $1.97 to $0.47 per run.
- Refutes asana.com:
On Model B, the best condition cut cost per run 29x and ran 4x faster than the original production setup.
- Context web.archive.org:
reduced estimated model cost from at least $36.21 (some original runs reached the step limit before they finished) to $1.24 per run
- Context asana.com:
Without batch pruning, caching the history at the larger budget cost more than not caching it on three of four models
- Context web.archive.org:
Each configuration performed the same task: collecting six fields for each of 32 books from a public demo catalog
- Context asana.com:
with three runs per condition: 144 runs, plus a 12-run follow-up. We calculated costs from each provider's token counters
- Context developers.openai.com:
Subsequent reads cost 0.1× that rate on most of these models and 0.05× on GPT-6.1 Sol
Open the Receipts Pack → What each source proves, every figure traced, and what would change our verdict.
