AI Model Speed, Measured: Tokens Per Minute in Real Sessions
How fast are Claude Opus 5, Fable 5.1 and Codex inside real agent sessions? Output tokens per minute from 118,896 measured requests, with the method and data.
A $5k audit PDF in three weeks that tells you what you already knew. A retainer split across thirty strangers. Shellharbour rankings checked quarterly, if at all.
The numbers on the right refresh with every deploy. Operator-led, plus agents, shipping weekly — your Shellharbour position compounds without anyone sending a Slack.
Shellharbour is one of the fastest-growing parts of the Illawarra, anchored by established suburbs like Warilla and Barrack Heights and the expanding Shell Cove development. The local economy leans on retail, trades, hospitality and healthcare serving a growing residential population. Demand is strongly local and "near me" driven, so Shellharbour campaigns need precise service-area targeting and fast, locally-relevant websites rather than broad metropolitan strategies.

Custom booking platform — calendar, payments, ops. Five times more leads per dollar and zero hours of weekly admin.

Suburb-page architecture + perf overhaul + Google Ads engineering.

Runtime-verified tracking + landing-page kit. Caught the silent drop a source-only audit missed.
Three things, run properly, forever improving.
Rank for the searches Shellharbour actually makes — technical, content and authority, run by agents.
Every dollar traced to a lead. Bids tuned daily, wasted spend cut weekly.
Sites and platforms that load instantly and convert. No WordPress, no page builders.
Illawarra — New South Wales. Local search, local intent, local results — campaigns tuned to how Shellharbour actually searches.















How fast are Claude Opus 5, Fable 5.1 and Codex inside real agent sessions? Output tokens per minute from 118,896 measured requests, with the method and data.
If your AI agent can edit the code that judges its own work, you don't have a quality gate — you have a suggestion. Ours proved the difference last week.
Our dashboard shipped nothing for 32 hours while the platform reported the same generic 'exit code 2'. Three stacked bugs — and one cause-agnostic fix.
Twilio says sent, your customer says nothing arrived. How Australian carrier filtering, sender ID registration and error 30007 actually work — plus the test protocol we run before blaming the platform.
A typical Australian service business runs five systems that never talk to each other. Here is what becomes answerable — cost per lead by channel, margin by service line, rebooking gaps — once they share data.
ServiceM8 is the best field tool in Australian trades, and it still can't tell you which channel your best clients came from. Here's what a proper API sync makes answerable, and why Zapier chains aren't it.
The same agent-run playbook, tuned to each market's own demand data — Shellharbour is one node of 661 we cover.