Skip to content

Enkidu · model inventory

Which coding model is good at what, and how fast.

Every model Enkidu confirmed it can run, rated per skill from published leaderboards (SWE-bench, Terminal-Bench, Design Arena, OpenRouter τ²-bench), ranked against every other model on the same board. Speed is the time to finish the same small coding task.

yellow = approximate: no result on our boards, so it takes the profile of the closest rated model from an announcement page's comparison table~ = one leaderboard onlyestimate from search= no benchmark number exists, so the standing is read off what search results say the model is (the quoted page); the weakest kind of number herefound by Arca research= a published figure Arca found on the web and verified on the cited pageupdated 2026-09-06 06:28 UTC · 21 models, 1 approximate
modelagentspeedoverallrepoterminalbackendfullstackfrontendalgotoolsgame3Ddata vizscithinkingprice
muse-spark-1.3-contributor-free
leaderboard
opencode
opencode
▮▮▮▯▯ 7.9s▮▮▮▮▮ 97~····99~··96~97~··minimal → xhighfree
opus
claude-opus-5
leaderboardfound by Arca research
claude
anthropic
▮▮▮▮▯ 6.9s▮▮▮▮▮ 97~100~100~10092~97~10099~97~96~98~·low → maxincluded in your claude plan
fable
claude-fable-5-1
leaderboardfound by Arca research
claude
anthropic
▮▮▮▯▯ 7.8s▮▮▮▮▮ 948869~100~9696~100~100100~99~100~·low → maxincluded in your claude plan
muse-spark-1.2-contributor-free
leaderboardfound by Arca research
opencode
opencode
▮▮▯▯▯ 8.0s▮▮▮▮▮ 90~86~86~·76~98~··94~92~98~·minimal → xhighfree
gpt-5.6-sol
leaderboardfound by Arca research
codex
openai
▮▮▮▮▯ 6.2s▮▮▮▮▮ 824950·82~98~·85~99~99~96~·low → ultra · default lowincluded in your codex plan
claude-fable-5
leaderboardfound by Arca research
claude
anthropic
▮▮▮▯▯ 7.4s▮▮▮▮▯ 79677280~95~94~0~100~97~94~97~·low → maxincluded in your claude plan
gemini-3.6-flash
leaderboard
opencode
google
▮▮▯▯▯ 8.7s▮▮▮▮▯ 79~···53~94~·61~84~89~94~·minimal → high$0.75 in · $3.75 out / M tok
sonnet
claude-sonnet-5
leaderboard
claude
anthropic
▮▮▮▮▮ 6.1s▮▮▮▮▯ 78~57~57~·87~87~·85~92~85~75~·low → maxincluded in your claude plan
gpt-5.5
leaderboard
codex
openai
▮▮▮▮▮ 5.9s▮▮▮▮▯ 74~90~90~·21~77~·75~93~67~80~·low → xhigh · default mediumincluded in your codex plan
gemini-3-flash-preview
leaderboard
opencode
google
▮▮▮▮▯ 6.1s▮▮▮▮▯ 7393~93~93~3562~·92~54~63~··minimal → high$0.5 in · $3 out / M tok
gemini-3.5-flash
leaderboard
opencode
google
▮▯▯▯▯ 23.7s▮▮▮▮▯ 7268~68~68~6181~·72~86~80~71~·minimal → high$1.5 in · $9 out / M tok
gemini-3.7-flash
leaderboardfound by Arca research
opencode
google
▮▮▮▮▮ 5.7s▮▮▮▮▯ 63~0~0~·68~92~·5495~94~95~·low → high$0.75 in · $3.75 out / M tok
gpt-5.6-luna
leaderboardfound by Arca research
codex
openai
▮▮▮▯▯ 7.8s▮▮▮▮▯ 637473··78~·49~64~45~62~·low → max · default mediumincluded in your codex plan
gpt-6-astra
found by Arca research
codex
openai
▮▮▯▯▯ 12.0s▮▮▮▮▯ 62~62~62~····88····low → ultra · default mediumincluded in your codex plan
gpt-5.6-terra
leaderboardfound by Arca research
codex
openai
▮▮▯▯▯ 9.6s▮▮▮▯▯ 574547·24~81~·64~75~50~77~·low → ultra · default mediumincluded in your codex plan
mimo-v2.5-free
leaderboardfound by Arca research
opencode
opencode
▮▯▯▯▯ 12.8s▮▮▮▯▯ 563319~40~·80~·51~78~70~79~·yesfree
nemotron-3-ultra-free
leaderboard
opencode
opencode
▮▯▯▯▯ 40.4s▮▮▮▯▯ 48~····36~·89~39~49~31~·yesfree
gpt-5.4-mini
leaderboard
codex
openai
▮▮▮▮▯ 6.6s▮▮▯▯▯ 34~······34~····low → xhigh · default mediumincluded in your codex plan
big-pickle
Arca's model identified it as GLM-4.6 (medium confidence) from https://grokipedia.com/page/Big_Pickle_model — "Community consensus identifies Big Pickle as the GLM-4.6 model from Zhipu AI, hosted under this codename by OpenCode."
estimate from search · medium confidence
grokipedia.com: “Community consensus identifies Big Pickle as the GLM-4.6 model from Zhipu AI, hosted under this codename by OpenCode.
opencode
opencode
▮▮▮▮▮ 5.0s332121212147·36504840·yesfree
nemotron-3.5-lightning-free
found by Arca research
opencode
opencode
▮▯▯▯▯ 19.3s▮▯▯▯▯ 88411~11~·······yesfree
gpt-5.3-codex-spark
codex
openai
▮▮▮▮▮ 5.3s···········low → xhigh · default highincluded in your codex plan

Index = the model's best published figure on each board, ranked 0–100 against every model on that board, weighted per skill. Hover a cell for the boards behind it. Rows come from the Enkidu inventory on one machine and refresh with it; a ≈ row is a stand-in until the model earns a row on a board we read.