Gemini 4 xhigh costs ~2.8x more but passes the same 12 checks as high in coding test
mhdfaran · x · 2026-11-23
Developer mhdfaran, building on SimonW's Gemini 4 Hydrogen thinking-effort comparison, ran his own test: an "interactive solar system in one HTML file" prompt executed twice at each level. All four runs passed all 12 functional checks (real orbital periods, Saturn's rings, the Moon, day counter, play/pause, speed slider, per-planet stats).
The differences were cost and polish, not capability:
- 10¢ → 28¢ per run (2.8x)
- 1.5 min → 4 min per run
- xhigh's extra thinking showed up as visual extras — label plates, textured Earth, richer info panels — with no additional requirements met
Takeaway: on this spec, the higher thinking budget bought visual niceties rather than better task completion. The author notes it's two runs on one task — a first read, not a rule. SimonW's original comparison was starker: high drew a pelican for 3.99 cents / 5,311 output tokens, xhigh for 23.61 cents / 31,477 tokens.
More from Models
- Gemini 4 Hydrogen at $1.50/M Input Tokens Scores 51, Beating Argon's 50 — dl_weekly · 2026-11-23
- antirez: don't let AI-skeptic programmers discourage you from the best tool devs ever had — antirez · 2026-11-23
- Base log perplexity is an extremely strong predictor of Epoch and OpenAI ECI, researchers note — stochasticchasm · 2026-11-23
- Redis creator antirez: AI is already great at math and programming, blaming the tool is the real problem — antirez · 2026-11-23
- Dylan Patel: efficient pretraining makes post-training misalignment exponentially easier — dylan522p · 2026-11-23
- Leaked OpenAI loss curve suggests GPT-6.2 Astra solved a Millennium Prize problem, internal model already lower — kimmonismus · 2026-11-23