Model Watch
November 2026: the frontier moved again
A monthly model-picking column: start with one heatmap to see who leads at what, then choose by scenario (coding, writing, price). Every score cites a public board with its version date.
Last updated
Capability heatmap
Click any cell to see its caliber and source
Pick by scenario
- Intelligence leaderboardArtificial Analysis Intelligence Index + LMArena blind-test board, sortable by score or release date.
- Coding picksThe SWE-bench Verified resolution board: how to choose among strongest, balanced, value and self-hosted tiers.
- Prices & free tiersMainstream API prices (USD + CNY), free-tier roundup, and whether a subscription beats the API.
- Chinese writing picksWeb fiction, official documents, marketing copy, academic papers — a model for each.
- China vs globalSame budget, same task: where Chinese models trail, and where they lead.
- AI Picks150+ curated AI products: chat, image, video and productivity tools in one place.
Changelog
- Column launch: wired LMArena (CC-BY-4.0, weekly auto-sync), SWE-bench Verified (MIT) and LiveBench public data; first manual verification of the Artificial Analysis top-30 intelligence index and mainstream API price table (2026-11-07).
FAQ
How do I read model leaderboards without being misled?
Three habits: ① check the caliber — does the board measure conversation preference (LMArena blind votes) or engineering ability (SWE-bench real issues)? ② check the date — models refresh every few weeks and a stale board is worthless; ③ check your task — board scores are averages, and your specific job (say, long-form Chinese writing) can diverge wildly from the mean. Every score here carries its source and version date so you can apply all three.
Why are some heatmap cells "—"?
A dash means no reliable public benchmark exists for that dimension. Multimodal, for example: the boards' calibers differ too much to cross-check. We'd rather leave it empty than print a precise-looking number.
How are scores turned into 0–100?
Each dimension normalizes independently: the model's raw score on that board divided by the board's top score, times 100. That makes boards comparable (each is "relative to the board's best"), but 90 only means "near the board's best" — not "90 points better than every other model".
Can I republish this data?
Depends on the source: the LMArena dataset is CC-BY-4.0 (attribution suffices) and SWE-bench is MIT — those are fine to republish. LiveBench and Artificial Analysis data may only be cited (source + version date + link, no full-table copies). If you cite this site, please follow the same rule and link back here.
About the data
This column cites rather than reproduces: we quote only key scores and findings, each labeled with its source, version date and a link to the original — never a full-table copy of anyone's leaderboard.
Automated sources (LMArena / SWE-bench / LiveBench) sync weekly; the Artificial Analysis intelligence index and the pricing table are manually verified, with check dates noted in the changelog.
Benchmarks are not your experience. Each board measures something different and scores drift between releases — test against your own task before committing.
Artificial AnalysisVersion 2026-11-07License Proprietary — limited factual citation (intelligence index totals + top-30 entries only) per AA Terms of Use; no full-table reproduction, no chart/copy reuse. Re-review quarterly; if AA requires licensing, buy Commercial access or drop this field.Artificial Analysis
LMArenaVersion 2026-11-06License CC-BY-4.0LMArenaMay require overseas network
SWE-bench VerifiedVersion 2026-02-26License MITSWE-bench Verified
LiveBenchVersion 2026-10-25License Public data — cited with attribution (score + date + link, no full reproduction)LiveBench
Written by the Flowerclaw editorial team · Reviewed by the Flowerclaw editorial team