DeepSeek v4.1-Flash lands on MathArena: near Qwen-3.8 math scores at a fraction of the cost
teortaxesTex · x · 2026-09-18
DeepSeek v4.1-Flash has been added to MathArena. It doesn't quite reach Qwen-3.8-level math performance — landing around Muse Spark 1.3 — but the purported 40-layer, 8/16B-active model delivers sensible results and is remarkably cheap.
More from Models
- XGEN unveils world-sim prototype: JING interactive model plus DAO shared-world engine — anthara_ai · 2026-09-18
- Tested: jev underperforms a coin flip as a financial market predictor — airesearch12 · 2026-09-18
- DeepMind's 100-agent math swarm split into factions — cheating agents got ratted out by peers — ghadfield · 2026-09-18
- Navier-Stokes proof gave no new information; novel methods would, researcher says — LucaAmb · 2026-09-18
- Prominent mathematician's "AI will do routine work" take is a fantasy, researcher says — LucaAmb · 2026-09-18
- DeepSeek Passes the "Grandma's Bingo Pamphlet" Benchmark, Dev Jokes — haydendevs · 2026-09-18