@julianharris@MaximeRivest Yes Sonnet 5 wasn't great but Sonnet 5.5 is very good.
Models are getting better and better quickly so it won't be long before we get much closer to frontier with local models.
@scheemunai I have been saying this to our clients for over a year. AI is the excuse for layoffs - most companies overhired during 2021-2022 and they are using AI to justify.
The best employees using AI are the last to get laid off as they are most productive.
Current Accounts 1 Claude $200 and 1 Codex $100 = $300
Big drop from the $800-1,000/mth we were spending. Local AI and Token Efficiency has helped us carve accounts down.
Did a complete redesign of the system that runs our main B2B business. We switched from a legacy vendor we were paying close to $10k a year.
Ours is AI native and does everything we ever wanted. Also built an accounting system from scratch to replace Xero/Quickbooks.
Ran Qwen 3.8 Next Flash locally on a 128gb machine and we were getting 73 t/s on numerous PRs we already merged as a test.
Quality wise, it's a step below Sonnet 5, and pales significantly with Sonnet 5.5
Still a great start.
@satellitedown@matt503ea5sf9z5 We just ran local tests on PRs we did to compare and Qwen 3.8 Flash is still below Sonnet 5 (not even close to 5.5) - we used a 4bit version.
2 of our sites were affected by a Google ranking swing in the 3rd week of October.
Sales have not been affected since we get a lot of traffic from Bing and LLM's - getting sales through Grok now too.
We weren't penalized - we will just have to wait it out a bit.
@SimonHoiberg BTW love your videos - much better these days with content.
Your videos a few years back had high production value but I much prefer cold hard facts than production.