Qwen3.8-Flash-Next

HOW WE PICK

• One board per job. Highest score among models that fit wins.
• Open models only, run with Ollama. We did not test these ourselves.
• Sometimes no model that fits here has an Arena score for a job. Sometimes a model without one beats the Arena pick on the Artificial Analysis index. Then we rank that whole job by the Artificial Analysis index instead, and say so on the SCORE line.
• A measured score beats an estimated score. An estimated score can be the pick if someone published a test on that computer for that job. It can also be the pick if no model with a measured score fits. The page then says that the score is an estimate.
• ADVANCED shows a higher-scoring model that only runs from a 4-bit or better file in llama.cpp, not as one Ollama command.
• Our judgment: under 15 tok/s is too slow for everyday use. Where a published test shows a model below that, it cannot be the main pick; it moves to an ALSO line.

FIT is our estimate from the published download size: Fits = size x 1.25 + 2 GB fits in usable memory; Tight = size + 1.5 GB fits. Usable memory: 192 GB Mac = 144 GB (our estimate). A tag with an open bug report for this kind of machine is never the RUN line; we use the standard tag instead.

• CHAT: Arena Text (overall), 2026-10-15
• CODING: Arena Text: Coding, 2026-10-15
• WEB APPS: Arena WebDev, 2026-10-19
• WRITING: Arena Creative Writing, 2026-10-15
• IMAGES IN: Arena Vision, 2026-10-19
• LONG DOCS: Artificial Analysis Intelligence Index. Must be Fits. No arena board exists for long documents.

Scores are the Artificial Analysis Intelligence Index, an independent test of overall quality (higher is better). Scores marked * (estimate) are estimates by Artificial Analysis, not measured results. CLOSE means the two scores' margins overlap, so the order could flip. Scores checked 2026-10-19.