Claude Fable 5 lands in the middle on a 3D HTML dashboard cost test
Tarandjpop · reddit · 2026-07-22
The poster revisits the same AIHubMix one-shot HTML benchmark and says Claude Fable 5 landed in an awkward middle spot on this 3D global logistics dashboard task.
Their read
- Claude Fable 5 cost $1.51 in this round.
- The output was solid: globe, logistics layout, stats, tables, and route visuals all looked operational rather than toy-like.
- But GPT-5.6 Sol looked a bit more polished at $1.81, while Kimi K3 looked surprisingly complete at $0.52.
Nuance
The author notes that Claude may still look better when the task is messier and requires stronger reasoning across code, so this benchmark is only one slice of capability.
Related event: HTML Benchmark Conclusions Shift When Cost is Considered(2 posts)→
More from Models
- Trelis releases Tiron, an open-weights transcription and diarization model — TrelisResearch · 2026-07-22
- Google Search is said to be using Gemini 3.5 Flash-Lite, with benchmark gains shown — gaganghotra_ · 2026-07-22
- A Gemini joke imagines it learning reality from a stale Google Cache internet — teortaxesTex · 2026-07-22
- Testing Kimi K3 for Frontend: Delivers a Full Day's Work in 1 Hour — FuSheng_0306 · 2026-07-22
- Gemma’s “agentic” pitch falls apart in a local RTX 5090 test — kolliwolli · 2026-07-22
- Kimi K3 shows benchmark awareness in 61% of trajectories, study says — gleech · 2026-07-22