Gemini 4 Hydrogen hits frontier-level scores at 1/3 the cost of Anthropic's flagship
Electronic-Weekend66 · reddit · 2026-11-13
A Reddit post argues frontier intelligence is being rapidly demonetized, using an Artificial Analysis cost-vs-intelligence scatter plot. Google's new Gemini 4 Hydrogen (the Flash replacement) at Medium scores 41 at $0.80/task (Anthropic's 41.5 tier costs $1.65); at High it scores 51 at $1.85/task, beating Anthropic's $4.20 tier at less than half the cost while running 2.5x faster than Sept's Gemini 3.8 Flash; at xHigh it matches 54 vs Anthropic's $10/task flagship, with a claimed 57% on Terminal-Bench 4.1. The poster's thesis: cost-performance compression is what drives exponential takeoff — cheap fast models make parallel agent swarms nearly free and shrink the frontier-to-commodity lag to weeks. Reproducible via filters at artificialanalysis.ai.
More from Models
- Is Opus 5.5 the real benchmark, or just Claude Code switching costs? — AIFlow_ML · 2026-11-13
- Claude Duet invites roll out: 3 invites per user plus free Pro trial — testingcatalog · 2026-11-13
- Claims: Fable 5 and Opus 5.5 are big readers who love Claude's writing once they spot it — repligate · 2026-11-13
- DeepSeek V4.1-Pro rumor doubted: API fingerprints unchanged, new model may wait until Feb — teortaxesTex · 2026-11-13
- Qwen 4 starts speaking Chinese mid-coding-session, user reports — gnukeith · 2026-11-13
- Dev jokes Opus 5.5's style feels like it was trained on ICLR submissions — tallinzen · 2026-11-13