Flash, Pro, Ultra → Autark: Why We Killed Our Own Product Tiers
We launched Autark with three tiers: Flash, Pro, and Ultra. Pick cheap, balanced, or premium. We pick the model. Simple, right?
Then we ran the numbers.
What thousands of evaluations taught us
We tested every model we serve against real business tasks — writing subject lines, analyzing financial models, drafting executive memos, qualifying sales deals, assessing compliance risk. Not academic benchmarks. Not trivia. The work our customers actually do.
Here's what we found: no model wins everything.
Different models are strong at different things. A model that excels at product strategy might struggle with financial analysis. The model that writes the best subject lines isn't the model you want drafting an investment memo. Routing every request to the same model, regardless of task, means leaving significant quality on the table.
The knowledge worker benchmark
Our users aren't latency-sensitive chatbots. They're knowledge workers — lawyers drafting contract analysis, product managers writing specs, sales teams researching accounts, analysts building financial models.
For these users, half a second of latency doesn't matter. Output quality does. A subject line that converts. A financial model that's correct. A compliance assessment that catches the edge case.
Traditional model benchmarks measure academic capability. They don't tell you which model writes the best investment committee memo or the most compelling sales follow-up. Our evaluations do.
So we killed the tiers
Flash, Pro, and Ultra were a proxy. A guess. "I want cheap" is not the same as "I want the cheapest model that's actually good at writing subject lines." "I want premium" is not the same as "I want the model that's proven best at pipeline analysis."
The new model is simpler:
- Autark — Smart routing. We identify what you're asking and route to the model proven best for that type of task.
- Autark Flash — Cost-optimized. The cheapest model that still delivers quality results for your task.
- Autark Deep — Deep reasoning. For complex, multi-step analysis where thoroughness matters more than speed.
Same API. Same endpoint. One line changes and you get the model that's actually best at what you're asking it to do.
What changes for you
If you're using flash, pro, or ultra today — they still work. We map them to the new modes behind the scenes. But you should switch to autark — the default. Your quality improves automatically as we evaluate new models and update our routing.
What's next
We're continuously evaluating models against new task types — legal, healthcare, finance — and expanding our benchmark coverage. We're also building tooling to let you run your own evaluations against your own quality standards.
Your model selection shouldn't be your problem. It's ours.