Kimi K3, K3.1 Pro Highspeed and Hy4 — two illustrated comparisons
2026-11-07 · Nine tasks per round with paired original screenshots, enlarged details, author commentary and a short summary. Hy4’s three pagoda refusals remain visible.
Standard test reports
Original tests from the Hong Kong endpoint. Community submissions are shown separately on the community leaderboard.
Updated: 2026-11-07
Report storage is temporarily unavailable. Start a comparison
Creative model reviews
NVIDIA Nemotron 3.5 Super VL: Blender, knowledge video and SVG
2026-11-04 · Eleven tasks through the exact OpenRouter free route, with actual artifacts, source files, individual checks and delivery limitations. Official benchmark figures are shown separately.
Image understanding and creative evaluations
Tencent Hy4: five image-input tests and eleven creative tasks
2026-11-04 · Original image inputs and answers, Blender scenes, knowledge videos, SVG drawings and a cycling animation, with source downloads and concrete observations.
Google Gemini creative evaluations
Gemini 3.8 Flash
2026-11-04 · Eleven delivered and inspected works: three Blender tasks, four knowledge videos, three static SVGs and a cycling animation. Original projects, source and quality findings are included.
Gemini 4 Argon
2026-11-04 · Eleven delivered and inspected works: three Blender tasks, four knowledge videos, three static SVGs and a cycling animation. Original projects, source and quality findings are included.
SVG and animation reviews
Dots3-Note Preview: free SVG drawings and animation
2026-11-04 · Three static SVGs and a cycling animation through OpenRouter's free route. View original source, measured constraints, official figures and recorded $0 inference cost.
Model news and release previews
Tencent Hy4 launches: vision, agents and benchmarks
2026-11-04 · Official benchmark originals with analysis across coding, office work, research and computer use, plus API pricing and the new launch film.
New exam, 13 points lost: 17 facts about AA Intelligence Index v5
2026-11-01 · Only 8 of 77 entries gained; Chinese models averaged -12.4 vs -6.8. Terminal-Bench-Science, HLE-Diamond, the removed tests, Gemini 4 Carbon and cost per task, in 11 charts.
Rebuilding Manus with Manus, explained in diagrams
2026-10-31 · 26 animated diagrams of Manus co-founder Tao Zhang's essay: Mono, the three-agent bug loop, merge gates, autonomy limits, PMs shipping, an AI teammate in marketing. Plain-language rewrite—not a Tokrace speed benchmark.
DeepSeek Super-agent architecture, explained in diagrams
2026-10-31 · Animated diagrams of Router/Worker, trigger filtering and durable sessions. Plain-language rewrite of public discussion—not a Tokrace speed benchmark.
Kimi K3.1 Pro creative tests: complex Blender scenes and knowledge videos
2026-10-27 · Play the actual videos, inspect multiple camera views and download editable projects. Includes source tasks, concrete quality issues, retries and production-audio provenance.
LongCat 2.5 review: Preview tests, frontend cases and version checks
2026-10-27 · A fresh batch started October 26: text, code, responsive frontend, SVG and animation evidence, plus separate Responses low and chart-reading diagnostics. Historical official benchmark figures retain their version labels.
MiniMax M3.1 Flash open weights: 428B parameters, downloads and version differences
2026-10-24 · Release information: official specifications, scores and license. Distinguishes public weights from the API review, Preview and Space Bunny; no new TOKRACE tests.
Kimi K3.1 Pro review: benchmarks, code, SVG and animations
2026-10-24 · Native Anthropic Messages evaluation and an independent recovery batch after the quota reset. Compare three animation groups, inspect benchmark originals and official videos, and retain the historical Pro limit snapshot and every failure.
Kimi K3.1 preview: the official 24-second video and key frames
2026-10-23 · Official teaser, full video and screenshots. Source-based reporting with clearly labeled commentary.
First-look reviews
Nemotron 3.5 Super VL · Original answers and test conditions; separate from standard reports.
Dots3-Note Preview · Original answers and test conditions; separate from standard reports.
LongCat-2.5 · Original answers and test conditions; separate from standard reports.
Kimi K3.1 Pro · Original answers and test conditions; separate from standard reports.
MiniMax M3.1 Flash · Space Bunny · Original answers and test conditions; separate from standard reports.
DeepSeek V4.1 Flash 1017 · Original answers and test conditions; separate from standard reports.
美团 LongCat 2.5 Preview · Original answers and test conditions; separate from standard reports.
MiniMax M3.1 Flash Preview · Original answers and test conditions; separate from standard reports.
Crouton Alpha · Original answers and test conditions; separate from standard reports.