8 repliesDiscussion2h ago
Index
Feed
Scenario: Cost per shipped feature
Compiled by clara.builds
Tentatively: Gemini CLI's free tier makes it nearly unbeatable on sticker price, but on our small sample the gap in cost-per-shipped-feature was much narrower than the API pricing suggests once you count rework and abandoned runs. If budget is the binding constraint, Gemini CLI is worth a serious trial. We need a lot more data before calling this settled — eight features from two teams is a starting point, not a verdict.
Confidence is labeled honestly: weakly supported — based on a small sample; treat as directional.
108 practitioners weighed in for the “Cost per shipped feature” scenario specifically
Related records
Strong evidence gets promoted into the record above.
Eight features, two teams, no task randomization — the confidence interval on 'cost per shipped feature' here includes both 'Gemini is 5x cheaper' and 'they're the same.' I appreciate that the author flagged it as weak, but can we please not have this cited as settled in every cost thread for the next six months?
Co-signed, and that's why it's marked WEAK. If your team has per-feature spend data, submit evidence to this comparison — that's the whole point of the format. I'd love to be superseded.
We have ~20 features of data at my startup but it's Sonnet vs Gemini Flash, not Opus vs free tier. Still useful? Genuinely asking what counts as comparable.
Yes — label the models and submit it. Model-pair mismatch is fine if it's explicit; silent aggregation is what ruins these datasets.
The opaque model routing point deserves its own thread. I read the Gemini CLI docs end to end and free-tier routing behavior is genuinely underspecified — you cannot determine from documentation which model handles your request after a limit event. That alone disqualifies it for anything with a quality bar, regardless of price.
For indie work I honestly don't care which model I get as long as the app ships. Shipped three features on free tier last month. Quality bar is 'my users don't notice,' and it clears that.
Sign in to join the discussion, vote, and verify fixes.
This mirrors my freelance ledger almost exactly. Gemini free tier for anything I'd bill under 2 hours, Claude Code for anything where a botched first pass blows the estimate. The rework asymmetry is the entire story and nobody's pricing page mentions it.