Model Kill Line
MODEL KILL LINE · AUGUST 2026
When a cheaper & stronger model makes a kill line
A model loses practical value when another model is both cheaper and more capable. Separate kill zones use DeepSeek V4 Flash 0731 and V4 Pro as references; their overlap marks models dominated by both.
V4 Flash blended price
$0.18 / 1M
DeepSeek overall score
1436
Models in kill zone
6
Current Pareto models
6
Overall capability vs. blended token price
Hover over any model for its LMArena Overall score, token prices, provider, and license. The x-axis blends 75% input and 25% output; DeepSeek uses converted official CNY cache-miss pricing and other models use OpenRouter pricing.
Providers
Killed by DeepSeek V4 Flash
Killed by DeepSeek V4 Pro
Pareto frontier
Reference model
Open weights (not OSI open source)
Proprietary
34 modelsHow to read this chart
01Higher is better. Overall scores come from LMArena Text Arena across math, coding, creative writing, and other open-ended domains.
02Further left is cheaper. Price is 75% input plus 25% output cost per million tokens; DeepSeek input temporarily assumes a 100% cache-miss rate.
03The red zone uses DeepSeek V4 Flash as its reference and the blue zone uses V4 Pro; their overlap means both models dominate that area.
Data sources
LMArena Text Arena — OverallOverall capability scores · leaderboard snapshot dated August 6, 2026OpenRouter Models APIInput and output token prices · retrieved August 8, 2026Official DeepSeek model pricingFlash: CNY 1 cache-miss input / CNY 2 output; Pro: CNY 3 cache-miss input / CNY 6 outputCFETS RMB central parity rateConverted at USD 1 = CNY 6.7928 on July 28, 2026