Qwen4-Max release
Qwen4-Max, previewed the previous Friday, is now a shipping model from Alibaba, according to Latent Space's AINews roundup.1 Artificial Analysis data cited by Latent Space show that at its highest reasoning setting the model scored 43 on the Intelligence Index, up from 34, while average cost per task fell from $10.02 to $3.27.23 Alibaba lowered its price for output from $6 to $4.50 for every million tokens, while the model generated roughly 39% less output.4
Alibaba kept total parameters at 2.4T but cut active parameters per token by 25% to 71B; the model accepts images and text and has a 1M-token context window.56 Smaller open-weight Qwen4 models have not yet been released but are still to come.7
According to Latent Space, Kimi K3.1 Pro had a cost of $3.68 for each index task and reached 67% on Terminal-Bench 4.0, compared with 48% for Qwen.89 Zvi Mowshowitz likewise notes that Qwen4 Max remains behind Kimi K3.1 Pro on Artificial Analysis.9 Mowshowitz judges that at a price of $1.50/$4.50 the model looks well below the cost-capability frontier.10
Sources quoted
Friday’s Qwen4-Max preview is now a shipping model.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026At the highest reasoning setting, Qwen4-Max scored 43 on the Intelligence Index, up from 34.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026Average cost fell from $10.02 to $3.27 per task.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026Alibaba cut output pricing from $6 to $4.50 per million tokens, and the model generated about 39% fewer of them.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026Alibaba kept the total parameter count at 2.4T and reduced the number active per token by 25% to 71B.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026The model takes images and text, with a 1M-token context window.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026Smaller open-weight Qwen4 models are still to come.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026Kimi K3.1 Pro cost $3.68 per index task and scored 67% on Terminal-Bench 4.0 to Qwen’s 48%.
— Latent Space, [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026Qwen4 Max exists, remains behind Kimi K3.1 Pro on Artificial Analysis.
— Zvi Mowshowitz, AI #195: Opening BellGiven the price of $1.50/$4.50 it seems well below the cost-capability frontier.
— Zvi Mowshowitz, AI #195: Opening Bell
Who reported it
- Zvi Mowshowitzrepeats Artificial Analysis: Alibaba's Qwen4 Max trails Kimi K3.1 Pro on Artificial Analysis. At $1.50/$4.50 it looks well below the cost-capability frontier.
- Latent Spacerepeats Artificial Analysis: Qwen4-Max moved from preview to release; Intelligence Index score rose 34 to 43 and cost per task fell from $10.02 to $3.27.
Facts and takes
Facts: 18 checked facts from 2 sources
- Alibaba released Qwen4-Max
- Qwen4-Max: $4.50 per 1M output tokens
- Qwen4-Max: 43 on Artificial Analysis Intelligence Index (highest reasoning setting)
- Qwen4-Max: 48% on Terminal-Bench 4.0
- Qwen4-Max: 1750 on Code Arena WebDev
- Qwen4-Max: 95.7% on HLAB (checklist items right)
- Qwen4-Max: 21.3% on HLAB (tasks passed)
- Qwen4-Max: 1M context window
- Qwen4-Max: output tokens generated down 39%
- Qwen4-Max: speed per image at high reasoning effort up ~4.5x vs Qwen3.8-max
- Qwen4-Max: cost per image at high reasoning effort down ~68% vs Qwen3.8-max
- Qwen4-Max: active parameters per token down 25%
- Qwen4-Max: output tokens up 3.6x
- Qwen4-Max: total parameters 2.4
- Qwen4-Max: active parameters 71
- Qwen4-Max: total parameter count 2.4T
- Qwen4-Max: active parameters per token 71B
- Qwen4-Max: accepts image and text input
Takes: 1 views from 1 holders · 1 negative
- -Roboflow :
Roboflow calls it underwhelming on its VLM benchmark
Timeline
- · Latent Space: [AINews] Perplexity hits a $1B revenue run rate, up 4x in 2026
- · Zvi Mowshowitz: AI #195: Opening Bell