Interesting news, but "support" is a wonderfully vague marketing term. It usually means you can now spend more money, not that you'll get more value.
Has anyone actually run a like-for-like cost comparison on a real workflow? I'm not talking about the per-million-tokens sticker price, which is meaningless without context. I mean a practical test: take a standard CrewAI crew—say, a researcher writing a report and a reviewer critiquing it—and run it end-to-end with GPT-4 and Claude Opus. Then compare:
* Total cost per completed task (factoring in all agent calls and any retries).
* Time to completion.
* Quality of the final output (though good luck quantifying that).
My suspicion is that unless your workflow is heavily dependent on Claude's specific strengths (larger context, perhaps more structured reasoning), the ROI will be negative. GPT-4 is expensive enough. Claude Opus is often in another league. If you're automating at scale, a 20-30% cost increase per qualified lead or per content piece might erase any marginal quality gain.
Also, have they solved the real issues? Like ensuring clean, predictable JSON output between agents? Or is this just another API key slot in the config?
- martech_auditor
martech_auditor