GPT-6.1 Terra costs half of GPT-6.1 Sol and trails it by four index points. A cost-per-task guide to choosing, routing and ...

One model fell from first to 69th without changing a weight. What crowd arenas, indexes and vendor sheets measure, and what ...

B total and 23B active is the short answer. The 252.4 GB download, the long-context concurrency and the token verbosity ...

Fable 5.1 and Mythos 5.1 share one set of weights and one price. What the safeguard tier changes, what access costs, and ...

Aion 3.5 Mini narrows the price gap to 4.7x, but caps output at 32K, takes no images, and has no published benchmark. Full ...

K context, native vision, GPQA 89.2. LBS-27B is an empty repo from the lab that built a probe on Qwen. What to watch.

Block's Dessa is full-duplex speech in and speech out, with no text step. What is published, what is missing, and where it fits a voice agent stack.

LFM2.5 24B A2B activates 2.45B parameters behind a $10M licence threshold. LBS-27B has no active-parameter count — and that ...

Gemma 4 26B A4B has 25.2B parameters, a 256K context and a live $0.06/$0.33 price. LBS-27B has a licence line. A dimension-by-dimension comparison.

DeepSeek V4.1 Flash is 13 times cheaper per token and 16 index points behind — but only 4.4 times cheaper per task. Where the cheap model wins.

Claude Opus 5.5 tops the Artificial Analysis index yet costs $5.98 a task. Why a 40% cost claim looks like a bubble signal, and what it means for buyers.

Claude Opus 5.5 is GA at $4/$20 with an AA Index of 58. DeepSeek V4.1 Pro has no card, price or weights. Compare the DeepSeek ...