Qwen4-Max benchmarks leak: 50% fewer tokens, 60% cheaper — but 'you've hit a wall'
teortaxesTex · x · 2026-11-16
teortaxesTex quotes a benchmark chart for a new MoE model claiming to beat Qwen3.8-Max with 50% fewer tokens and 60%+ lower cost per task, and argues that in late 2026 such gains only qualify as a next-gen FLASH-tier model, not a flagship — "you've hit a wall." The attached chart shows Qwen4-Max leading Qwen3.8-Max, Opus5.5, GPT6.1-Sol, GPT6-Astra and Kimi K3.1 across DeepSE, CodeArena, Terminal-Bench 4.0, JobBench, Humanity's Last Exam and more.
Related event: Qwen4-Max benchmark charts leak: half the tokens, over 60% cost cut(2 posts)→
More from Models
- Qwen4 Max per-eval breakdown: strong on knowledge work, lagging in coding — ArtificialAnlys · 2026-11-16
- Qwen4 Max scores 43 on AA Intelligence Index, 67% cheaper per task than predecessor — ArtificialAnlys · 2026-11-16
- Qwen4-Max debuts at #14 on Agent Arena, making Alibaba the #5 lab overall — Alibaba_Qwen · 2026-11-16
- New MoE base claims win over Qwen3.8-Max with 50% fewer tokens, met with skepticism — teortaxesTex · 2026-11-16
- Comparing AI censorship: Gemini preempts, Claude silently rewrites, OpenAI just blocks — ivan_bezdomny · 2026-11-16
- Qwen4-Max lands #6 on Code Arena WebDev, up 74 pts over Qwen3.8-Max — Alibaba_Qwen · 2026-11-16