100 models and 8 new vendors
Steven · 2026-08-20
We just crossed 100 models on the table, across 19 vendors. Here's everything new.
FEAT8 new vendors joined: Cohere, AI21 Labs, NVIDIA, Amazon, MiniMax, ByteDance, Baidu, and Tencent.
FEATCohere brings Command A (command-a), AI21 brings Jamba Mini and Jamba Large, and Amazon brings the full Nova lineup: Micro, Lite, Pro, Premier, and the new Nova 2 Lite.
FEATMiniMax (minimax-m2.7), ByteDance's Doubao Seed 2.1 Pro, and Tencent's Hunyuan HY3 round it out, alongside Baidu's ERNIE 5.1 and ERNIE 4.5 Turbo. ERNIE 5.1 steps up to a higher rate once a request crosses 32K tokens.
FEATNVIDIA's Nemotron 3 Ultra has no metered price at all. The table now shows a Free tag for models like it, separate from the Plan tag used for subscription-only access.
FEATGoogle added Gemini 3 Flash Preview (gemini-3-flash-preview), $0.50 input / $3.00 output.
FEATOpenAI added GPT-5.6 Cyber (gpt-5.6-cyber), a cybersecurity-research model gated behind the Daybreak Red program. $12.50 input / $75 output, the most expensive model on the table right now.
FEATGLM-5.3 has real per-token pricing now, $1.40 input / $4.40 output, after launching without one. It's still bundled into the GLM Coding Plan as well.
FEATDeepSeek split V4 Pro and V4 Flash into peak and off-peak billing. Off-peak, which covers most hours, is what's shown: $0.66/$1.98 and $0.22/$0.66. Peak hours, 01:00-04:00 and 06:00-10:00 UTC, bill at exactly double that.
FEATPerplexity's whole Sonar line picked up a notice: it retires on 2026-09-27 in favor of the new Agent API.
FIXClaude Opus 4 now shows as retired. It had been mismarked as deprecated.
FIXGPT-4.1 had been mismarked as deprecated too. Fixed.