Pelican test on MiMo 2.6: plan mode cuts agent calls to 12-17 vs heavy iteration
PilgrimofHaqq2 · reddit · 2026-09-28
A controlled pelican SVG test comparing MiMo 2.6 (Flash/Pro) with and without plan mode:
- Plan mode: settles style/scene/size in one Q&A round, writes the full SVG in a single call (11.1KB Flash, 22.1KB Pro), and batches all render fixes into one edit — just 12 and 17 total calls.
- No plan: far more iteration. Flash needed 3 render-fix rounds and lost 10 calls to image verification (mismatched crop reads, zoomed views, one stale preview); Pro did 2 fix rounds plus 4 tool mishaps, one generating 3,743 tokens that were thrown away when the edit call was rejected for a missing argument.
- Surprise finding: Flash with plan generated more tokens than without (27.2k vs 20.4k) — fewer, bigger calls doesn't mean less work.
Takeaway: plan mode's value is consolidating fragmented iteration into a few large calls, not reducing total token spend.
More from coding & agent
- Jev seen as fit for high-level robotics orchestration: fast schema beats LLM lag — ai · 2026-09-28
- Apple's free on-device fm paired with decision model Jev beats big-model routing in tests — jasonkneen · 2026-09-28
- Codex builds whole apps but keeps choking on reusing your live Chrome session — ChrisGPT · 2026-09-28
- DeepSeek ships Codex-style agent harness that connects to multiple model providers — ChrisUniverse · 2026-09-28
- Indie dev builds CoIsland, a notch-resident AI monitor with 37 connectors, entirely with Claude Code — DonTizi · 2026-09-28
- Andrew Carr says Astra gave bit-identical, hash-verified outputs on his weekend project — andrew_n_carr · 2026-09-28