Kimi K3.1 Pro costs $1.98/task on the Artificial Analysis Intelligence Index vs $5.98/task for Claude Opus 5.5 (score 55 vs 58).
On Terminal-Bench 4.0, Kimi hits 67%, beating Claude Sonnet 5.5 (64%).
Pay 3x for Opus, or switch to open weights?
On the Artificial Analysis Coding Agent Index v2.0, Codex with GPT-6.1 Sol scores 54.1 at $1.50/task.
Codex with GPT-6 Astra scores 54.7 at $7.04/task.
That's 4.7x the API bill for 0.6 points.
Pay 4.7x for half a point, or pocket the $5.54/task?
A 2015 image model just beat two 2026 frontier LLMs at skin cancer diagnosis.
Fine-tuned Inception v3: 125 of 200 biopsy-proven lesions right (62%). GPT-6 Astra and Claude Fable 5.1: 84 and 85 (42%).
Medical imaging: small specialist model, or wait for generalists?
Upon arriving at the museum, I discovered a truly unique and albeit slightly terrifying object. I decided to visit after the sun had set, and to be honest, it took me a bit of time...


