Qwen4 Max lands on Together AI at 30% off: 2.4T-param MoE with 1M context

togethercompute · x · 2026-11-20

Alibaba's flagship Qwen4 Max is now live on Together AI with a 30% discount.

The model is a sparse MoE with 2.4T total parameters and 71B active—Alibaba reports 25% fewer active params than Qwen3.8 Max at the same scale—built on the new Qwen4 architecture with upgraded attention, residual connections, and embeddings. It offers a 1M-token context window, native vision input, deep thinking enabled by default, and targets long-running agentic work across software engineering, finance, law, and research. Pricing: $1.05 per 1M input tokens, $3.15 per 1M output tokens.

Original post →

More from Infra

Infra channel →