Qwen3.8-Max
Alibaba · released
Alibaba's top hosted Qwen: a 2.4-trillion-parameter model that takes text, images and video with a 1M-token context, aimed at long autonomous coding, office work and multimodal agents. It has the highest Arena Vision rating of any model tracked here.
Best for: Image and video understanding · Long coding sessions · Multimodal agents
- Intelligence Index
- 45.4
- #10 of 38
- Price per 1M tokens
- $2 in $6 out
- Blended $3 at 3:1
- Context window
- 1M
- Up to 131K output
- Output speed
- 37 tokens/s
- $5.41 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall69 · #12 of 38
Based on 3 of 3 benchmarks
- Coding56 · #16 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work68 · #15 of 40
Based on 6 of 6 benchmarks
- Reasoning & science60 · #17 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability70 · #16 of 39
Based on 2 of 2 benchmarks
- Vision89 · #5 of 31
Based on 2 of 2 benchmarks
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#10 of 38· Artificial Analysis
- Vals Index48.3%
#20 of 26· $4.96 per test· Vals AI
- Arena · Text1482 ±5
#9 of 33· Max effort· 23,353 votes· Arena
Coding
- Terminal-Bench 4.038.9%
#8 of 38· Artificial Analysis
- SciCode52.1%
#19 of 38· Artificial Analysis
- Arena · WebDev1672 ±12
#7 of 34· Max effort· 3,450 votes· Arena
- Vibe Code Bench64.7%
#22 of 26· Vals AI
- Code Migration23.2%
#20 of 26· Vals AI
Agents & knowledge work
- GDPval-AA58.6
#7 of 39· Artificial Analysis
- Arena · Agent+2.5 ±0.6
#14 of 29· Arena
- Finance Agent v250.6%
#20 of 26· Vals AI
#15 of 26· Vals AI
- Legal Research Bench47.6%
#7 of 26· Vals AI
- Tax Agent Bench66.0%
#9 of 26· Vals AI
Reasoning & science
- Humanity's Last Exam43.1%
#14 of 39· Artificial Analysis
- CritPt17.7%
#17 of 39· Artificial Analysis
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA Diamondsaturating92.8%
#8 of 27· Artificial Analysis
Knowledge & reliability
#14 of 39· Artificial Analysis
- AA-LCR80.3%
#20 of 39· Artificial Analysis
Vision
- MMMU-Pro82.8%
#5 of 27· Artificial Analysis
- Arena · Vision1301 ±7
#1 of 19· Max effort· 10,040 votes· Arena
Reported by Alibaba
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- SWE-bench Pro · Claude Code harness67.7%
- OSWorld-Verified86.1%
- MMMU-Pro82.3%
- GPQA Diamond92.6%
Specifications
- Lab
- Alibaba
- Released
- Aug 3, 2026
- Availability
- Generally available
- Context window
- 1,000,000 tokens
- Max output
- 131,072 tokens
- Accepts
- Text, Images, Video
- Knowledge cutoff
- Not published
- Weights
- Closed
- Parameters
- 2.4T total · 95B active
- Reasoning control
- Effort low to xhigh · can be switched off
- API model ID
qwen3.8-max
Pricing
- Input / 1M tokens
- $2
- Cached input / 1M
- $0.25
- Output / 1M tokens
- $6
- Blended (3:1)
- $3
- Measured cost / task
- $5.41
International list price on Alibaba Cloud Model Studio.
Worth knowing
- A text-only version of the same model is downloadable as Qwen3.8-2.4T-A95B under a custom licence.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.