Alibaba’s Qwen 3.8 Max Prime is now on MiniRouter. It serves the Qwen 3.8 Max model with 1.5 to 2x the output speed, at twice the token price.
Fireworks’ Ember 1 joins it. Built on Kimi K3 and trained to skip reasoning that does not change the answer, it reports Kimi K3’s quality with 40% fewer tokens and 82.0% on Terminal Bench 2.1. MiniRouter did not run these evals.
Three models
- Qwen 3.8 Max PrimeVercel list: $4 in, $12 out per 1M tokens. Same model, served faster.alibaba/qwen3.8-max-prime
- Qwen 3.8 MaxVercel list: $2 in, $6 out per 1M tokens. The standard lane.alibaba/qwen3.8-max
- Ember 1Vercel list: $3 in, $15 out per 1M tokens. Kimi K3 quality with 40% fewer tokens, per Fireworks. Research preview.fireworks/ember-1
Try it
All three take 1M tokens of context, text and images, tools and JSON output. Send as the model.
curl https://api.minirouter.sh/v1/chat/completions \
-H "Authorization: Bearer $MINIROUTER_KEY" \
-H 'Content-Type: application/json' \
--data '{
"model": "alibaba/qwen3.8-max-prime",
"max_tokens": 1024,
"messages": [{ "role": "user", "content": "Fix this failing test." }]
}'Benchmarks and prices: Qwen3.8-Max release post, Fireworks’ Ember-1 post.