GPT-6 Astra costs $50 per million output tokens vs DeepSeek's $1.20 — frontier intelligence is cheap if you route tasks right
alex_verem · x · 2026-09-22
In his article "frontier intelligence is actually cheap," the author argues that using one AI model for everything is why people burn through usage limits by lunch or get shallow answers on questions that matter.
- Quick jobs (rewriting emails, summarizing, translating, fixing formulas) only need speed — small fast models like Gemini 3.8 Flash or DeepSeek-V4.1-Flash match frontier quality. DeepSeek charges $1.20 per million output tokens vs $50 for GPT-6 Astra; the extra $48.80 buys nothing on these tasks
- Thinking jobs (decisions, contracts, plans, long documents with buried contradictions) are where a wrong answer costs something and frontier models earn their price
The takeaway: frontier intelligence is cheap in aggregate — the real waste is spending it on tasks that don't need it.
More from Models
- User presses Claude Code lead on whether usage resets will actually be banked — Angaisb_ · 2026-09-22
- Anthropic's 'Constitution' Is Just RLAIF, Argues Researcher — Its Real Benefit Is Cutting Human Labeling — WillRinehart · 2026-09-22
- Qwen4 training now: Alibaba teases Max/Flash/Plus variants, 5-10T params for Qwen4.5-5 — ChrisGPT · 2026-09-22
- Reverse-engineered Splash format ports Qwen3.8-27B to Mac, 88 tok/s at 64k on M5 Max — SeveralViolins · 2026-09-22
- Fable can interrupt itself: answer your new prompt, then resume the first — gleech · 2026-09-22
- Did OpenAI Solve the Wrong Navier-Stokes Problem? Experts Cry Loophole — joshgans · 2026-09-22