Stop defaulting to Medium: benchmarking Gemini 4 Argon vs 3.8 Flash effort budget token costs
rseroter · x · 2026-10-16
A post argues against lazily defaulting to Medium effort levels in Antigravity, sharing token-usage data for Gemini 4 Argon and Gemini 3.8 Flash across four coding tasks (env audit, bool refactor, linear backoff feature, CLI dry-run).
Key takeaways from the charts (averaged runs, split by cached input/input/output tokens):
- Gemini 3.8 Flash: High effort is dramatically more expensive — up to 2.5M–5.7M tokens on some tasks; Medium is second; Low is cheapest
- Gemini 4 Argon: an order of magnitude cheaper overall, mostly under 0.5M tokens at every level, and Medium is sometimes cheaper than Low
The lesson: effort budgets affect cost very differently per model and task type, so it's worth choosing levels deliberately rather than defaulting to Medium.
More from coding & agent
- raylib 6.2 ships with 300+ improvements and 100 new contributors, 20 new API functions — jsuarez · 2026-10-16
- "Vibecoding" is as misleading a label as "AI" and "SI," dev complains — eschadiol · 2026-10-16
- Dev runs Pstack as a Grok bot to audit his own pipeline — it surfaces a pile of hidden bugs — alexcovo_eth · 2026-10-16
- Doc2Harness: New benchmark finds best coding agent passes only 53% across 329 cases — _akhaliq · 2026-10-16
- Perplexity open-sources SPACE, the Firecracker microVM sandbox behind Perplexity Computer — perplexity_ai · 2026-10-16
- Blackstone goes all-in on Factory for autonomous software development — matanSF · 2026-10-16