Plan guide

How much coding do you get from Codex Go?

Available models: GPT-6 Astra / GPT-6.1 Sol / GPT-6 Luna. $8/mo buys about 2 h/week of building at p50 peak (p10–p90 1–4) — hours are guessed, so treat them as a guess. Coding-agent quality scores (/100): GPT-6 Luna (max) · Codex · 41/100 planning, GPT-6 Luna (max) · Codex · 41/100 building, GPT-6 Luna (max) · Codex · 41/100 admin. Building model cost: $0.18 per benchmark task at API prices. Catalog note: ChatGPT Go ($8): Codex in the desktop app only — no CLI, IDE or cloud tasks — running GPT-6 Luna at Standard speed. Hours set below Plus; the seat is for lightweight tasks, per OpenAI. Guessed — no lived reports yet.

Hours by role

p10 / p50 / p90, h/week, at the plan's own clock. Peak and off-peak differ only where the vendor publishes a night rate.

RoleBenchmark quality /100Peak p10 / p50 / p90 (h/week)Off-peak p10 / p50 / p90 (h/week)
PlanningGPT-6 Luna (max) · Codex · 41/1001 / 1 / 2—
BuildingGPT-6 Luna (max) · Codex · 41/1001 / 2 / 4—
AdminGPT-6 Luna (max) · Codex · 41/1002 / 4 / 8—

Clock: no published off-peak discount. No CLI agent. Source: guessed.

Get Codex Go →Rank it against your stack →

Current coding-agent evaluations

The ranking uses the fixed setting named in each role above. These scores do not establish subscription hours. Model availability verified 2026-11-08: vendor documentation.