Replotting coding-agent costs by API price flips the picture, and OpenAI devs' median daily spend was near $0
eliebakouch · x · 2026-09-07
- Discussing Astra's coding-agent usage stats, eliebakouch notes the original chart's y-axis isn't normalized by API retail price, which inflates the apparent impact of the cheaper Sol; he shares a replotted version that accounts for pricing.
- He cautions that token consumption alone is a poor cost metric: pricier per-token models may use fewer tokens while delivering more intelligence.
- A surprising data point: the median daily spend per researcher at OpenAI in February 2026 was close to $0, even though OpenAI's best public model at the time was GPT 5.3 Codex.
- dbreunig doubts promotions were included in the pricing, points out Sol's relentless persistence doesn't show up on the chart, and agrees pricing is the key variable.
More from coding & agent
- Perplexity goes local: private tasks hand off to on-device small models — HowDevelop · 2026-09-07
- Claude Code team interview: delegating whole projects, not just functions — xiaohu · 2026-09-07
- Undocumented Codex 'remote-control' feature lets you drive GPU clusters from your phone — iScienceLuvr · 2026-09-07
- An MCP server just for telling the time: Stamp fixes LLM's frozen-clock gap — paulahjort · 2026-09-07
- OpenAI's Jeff Weinstein: a chief-of-staff bot delegating to domain sub-bots is very powerful — jeff_weinstein · 2026-09-07
- Memory, not models, is the moat: AI must know what changed, not just what was said — rohanpaul_ai · 2026-09-07