Atlas

Models

← All models

Qwen3.8 27B

Alibaba

Qwen3.8 27B is Alibaba's dense, post-trained vision-language checkpoint for coding and agentic work, accepting images and videos. It supports optional thinking with low, medium, and xhigh effort, and a native 262,144-token context extendable to 1,000,000 with the documented scaling configuration.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo1398
AA-Briefcase v1.1 Elo1403
AA-LCR v1.182.0
AA-Omniscience Index-8.0
Arena Code WebDev — Brand & Marketing1614
Arena Code WebDev — Consumer Product1571
Arena Code WebDev — Content Creation Tools1537
Arena Code WebDev — Data & Analytics1548
Arena Code WebDev — Gaming1637
Arena Code WebDev — HTML1579
Arena Code WebDev — React1600
Arena Code WebDev — Reference-Based Design1605
Arena Code WebDev — Simulations1586
Arena Image-to-WebDev1582
Arena Text — Business, Management & Financial Ops (No Style Control)1440
Arena Text — Chinese (No Style Control)1500
Arena Text — Creative Writing (No Style Control)1378
Arena Text — English (No Style Control)1445
Arena Text — Entertainment, Sports & Media (No Style Control)1386
Arena Text — Exclude Ties (No Style Control)1434
Arena Text — Exclude Ties (Style Controlled)1433
Arena Text — Expert (No Style Control)1491
Arena Text — French (No Style Control)1462
Arena Text — French (Style Controlled)1463
Arena Text — German (No Style Control)1447
Arena Text — Hard Prompts (No Style Control)1455
Arena Text — Hard Prompts English (No Style Control)1454
Arena Text — Instruction Following (No Style Control)1436
Arena Text — Japanese (No Style Control)1370
Arena Text — Korean (No Style Control)1402
Loading Atlas data…