Latest changes
Changes to boards, recommendations, and methods. The scores themselves stay on the scenario pages.
10-29BoardTwo-board sync: LMArena WebDev to Oct 27 (952,039 votes, 153 rows); Opus Max still #1 (1810); new Gemini 4 Carbon (high) at #4 (1768, preliminary). EQ Creative Writing 146 rows; Astra still #1 (Elo 2176.9); Kimi K3.1 Pro enters at #2 (2172.6).lmarena.ai / eqbench.com
10-28BoardSix-board sync: sudo L7 16 rows, Sonnet 5.5 leads at 50.6%; Kimi K3.1 Pro ties Qwen at #6. WebDev to Oct 27 (950,329 votes, 152 rows); Opus Max still #1; new GPT-6.1 Terra at #14. After Slop inversion Inkling Small leads (76); Haiku about #4 (old “Haiku #1” stale). Android Bench 2.0 11 rows, Sonnet max 40.0% leads. FrontierCode upgraded to 2.0 (102 rows / 26 models; complete JSON includes Obsidian 3). Convex No/With Guidelines refreshed; With Guidelines Argon ~98.7%.surgehq.ai / lmarena.ai / theslopindex.com / developer.android.com / cognition.com / convex.dev
10-26BoardFour-board sync: LMArena WebDev to 2026-10-22 (923,854 votes, 151 rows); new Moonshot Kimi K3.1 Pro (max) at #8 (1692, preliminary). Slop Index expands to 58 rows (2,091 votes); after inversion Haiku 5.5 leads (100−17=83). EQ Creative Writing 145 rows; Meta muse-wonder at #9 (Elo 2007.0). Convex With Guidelines refreshed; Claude Opus 5 at #9 (94.64%).lmarena.ai / theslopindex.com / eqbench.com / convex.dev
10-20BoardFour-board sync: LMArena WebDev to 2026-10-19 (903,893 votes, 149 rows); Opus 5.5 (max) leads at 1807; new Muse Wonder Full (max) at #6 (1726). Image-to-WebDev to 2026-10-19 (191,568 votes, 65 rows); Opus 5.5 (max) leads at 1753; DeepSeek-V4.1-Flash-20260910 (max) at #13 (1606). Convex No / With Guidelines have 38 models each; new GPT-5.5 at #14 (77.68%) / #17 (89.29%).lmarena.ai / convex.dev
10-19BoardThree-board sync: LMArena WebDev to 2026-10-19 (900,155 votes, 146 rows); Opus 5.5 (max) leads at 1806; new DeepSeek-V4.1-Flash-20261017 (max) at #12 (1670). ReactBench v1.2 resynced (85 configs) with Opus 5.5 (max) 54.4% at #1. Convex adds With Guidelines and refreshes No Guidelines / Web.lmarena.ai / reactbench.com / convex.dev
10-18PickReact Bits: refreshed the Motion one-liner from the site (free open-source animated React components, copyable source), plus repo and a shadcn install example.reactbits.dev
10-14BoardLMArena WebDev: synced to the 2026-10-13 board (878,407 votes, 143 rows). Claude Opus 5.5 (max) still leads at 1811; new Meta muse-wonder-max at #15 (1651).lmarena.ai
10-13BoardFrontierCode 1.1: 130 Main rows (all reasoning levels). Claude Opus 5.5 54.6%. General board on Web frontend / Android / CLI at weight 0.6.cognition.com/frontiercode
10-12BoardWrite docs & copy: adds EQ-Bench Creative Writing v3, 140 Elo rows; weight 0.6 (Slop Index stays 1.0). Composite lead is GPT-6 Astra.eqbench.com
10-12PickBest Free Fonts: added to Type—a curated free-font directory focused on truly free faces, commonly SIL Open Font License.bestfreefonts.com
10-11BoardDeepSWE v1.1: 28 Best-config rows. GPT 6 Astra 74.1%, Gemini 3.8 Flash 73.8%, Claude Opus 5 73.6%. General board on Web frontend / Android / CLI.deepswe.datacurve.ai
10-11BoardWrite docs & copy: new scenario on The Slop Index, 39 rows. Source ranks sloppiest first; after inversion GPT-5.6 Luna leads (100 − Slop = 74).theslopindex.com
10-09BoardLMArena Image-to-WebDev: 64 models. Claude Opus 5.5 (max) scores 1749. This is image-to-page, not the text-prompt WebDev board.arena.ai
10-09Boardsudo L7: Claude Opus 5.5 and Claude Sonnet 5.5 both score 45%, with Opus listed first. GPT 6 Astra scores 32.2%.surgehq.ai
10-08BoardLMArena WebDev: Claude Opus 5.5 (max) leads at 1813, 27 points ahead of GPT-6 Astra.lmarena.ai
09-29BoardTerminal-Bench 4.0: GPT-6.1 Sol ties for 3rd at 58.2%, at $634.24.tbench.ai
09-29PickThe full-stack set uses GPT-6.1 Sol from Terminal-Bench: best once cost is included.编辑
09-22BoardTerminal-Bench 4.0: Claude Opus 5.5 takes first at 64.8%.tbench.ai
08-11BoardAndroid Bench: adds eight models, including Claude Opus 5, GPT 5.6 Sol / Luna / Terra, Kimi K3, and Qwen3.8 Max.android.com
—MethodSWE-bench Verified has not listed a new model generation since February 2026, so it is not primary evidence.编辑