Qwen-4-Flash-Early tops GameFeel Bench at 49.9, ~7.6x cheaper than rivals
pbaylies · x · 2026-11-08
GameFeel Bench v1.4 adds Qwen-4-Flash-Early (codename hawk), which takes the overall top spot at 49.9 — ahead of Sonnet-5.5 (47.2), Opus-5.5 (45), and sibling Qwen-4-27B-Early (44.6).
- On the price-to-performance frontier, Qwen-4-Flash-Early ($0.40/task) lands 7.6x ahead of its cost; Qwen-4-27B-Early (High) is 8.4x cheaper than expected at $1.34/task
- Subcategory charts show strong Gamefeel-Re results for the Qwen-4 pair
- The maintainer also fixed a genuine miscalculation in the earlier qwen-4-27b audio score
Both Qwen-4 models carry "Early" labels, i.e. pre-release versions, so treat the numbers as provisional.
More from Models
- Polymarket gives Claude 52% to be best AI by end of 2026, ChatGPT just 9% — nian2326076 · 2026-11-08
- AI Daily: Anthropic's hidden saboteur AI evades teammates in EVE Online experiment — xiaohu · 2026-11-08
- Reddit users report sharp model regression: hallucinations, sycophancy, flip-flopping — DanceTheNight88 · 2026-11-08
- GLM-5.3-SLIM GGUF quantization trends on Hugging Face for llama.cpp — Scantist · 2026-11-08
- Claude Pro plans have no 5-hour limits: 10x to 50x Plus usage clarified — TheMoonMidas · 2026-11-08
- New Chimera voice model one-shots the shoebody bop meme almost perfectly — theshawwn · 2026-11-08