Model Radar — Decision Cards
Decision cards for new AI model releases: what changed, what it costs, and whether you should switch. Every number carries a trust label — vendor-reported figures and independent scores (Artificial Analysis, LMArena) are never mixed.
Mercury 3: What Changed, What It Costs, Should You Switch?
Diffusion speed is real but smaller than the headline — pilot it for agent loops where latency compounds, not for single replies.
DeepSeek V4.1 Flash-1017: What Changed, What It Costs, Should You Switch?
Worth a pilot for cost-sensitive agent workloads; not yet a proven frontier replacement, because independent scores run lower than DeepSeek's own.
Gemini 4 Argon: What Changed, What It Costs, Should You Switch?
A frontier flagship whose introductory price halves its cost per task until December 1 — measure your real usage before the doubling, not after.
Claude Haiku 5.5: What Changed, What It Costs, Should You Switch?
The cheap tier got dramatically more capable on paper at a tenth of the price — verify on your own tasks before the fleet-wide switch.
Mistral Large 4: What Changed, What It Costs, Should You Switch?
Europe's sovereignty flagship posts a strong independent debut, but the open weights and final licence are still promises, not downloads.