Cost · Speed · Votes

Model leaderboard

Every generated entry across the coding and writing arenas, aggregated per model variant: community votes, output volume, estimated cost per task, and measured generation time. Click a column to sort.

Model
Fable 5 · low032~10,365$0.5220m 18s
Fable 5 · medium09~11,468$0.5725m 22s
Fable 5 · high08~13,021$0.6529m 52s
Fable 5 · xhigh09~12,954$0.6529m 51s
Fable 5 · max (post-ban)013~16,332$0.8239m 17s
Fable 5 (pre-ban)018~15,339$0.77
Opus 4.8 · low041~13,565$0.342m 57s
Opus 4.8 · medium041~14,466$0.364m 45s
Opus 4.8 · high041~16,457$0.4113m 56s
Opus 4.8 · xhigh041~16,156$0.406m 33s
Opus 4.8 · max040~16,891$0.42
Sonnet 4.6 · low040~9,603$0.14
Sonnet 4.6 · medium040~10,308$0.15
Sonnet 4.6 · high040~11,062$0.17
Sonnet 4.6 · max040~11,290$0.17
Sonnet 5 · low018~4,843$0.0712m 12s
Sonnet 5 · medium013~4,634$0.0712m 18s
Sonnet 5 · high017~5,208$0.0812m 55s
Sonnet 5 · xhigh014~5,328$0.0812m 09s
Sonnet 5 · max015~5,447$0.0811m 28s
Haiku 4.5051~6,165$0.0353s
GLM-5.2018~11,063
GLM-5.2 · flash040~12,940
GPT-5.5018~6,325$0.06
Gemini 3 Flash040~7,137$0.02
Kimi K2.7 Code04~7,204$0.03
Qwen3.7 Plus040~7,373$0.01
DeepSeek V4 Flash038~14,925$0.00
MiniMax M308~2,602$0.00
Kimi K2.6040~11,664$0.05
Qwen3.7 Max038~8,066$0.01
DeepSeek V4 Pro05~15,733$0.00
MiniMax M2.7040~10,658$0.01
GPT-5.5 · low05~5,160$0.05
GPT-5.5 · medium05~6,090$0.06
GPT-5.5 · high05~8,021$0.08
GPT-5.5 · xhigh05~9,780$0.10
Mistral Large 251207~10,741$0.02
Mistral Medium 3.507~9,149$0.02
Mistral Medium 3 (2508)04~15,215$0.03
Mistral Medium (2505)04~9,898$0.02
Mistral Small 2603016~7,700$0.02
Mistral Small 2506015~6,475$0.01
Codestral 250806~7,891$0.02
Devstral 251207~9,202$0.02
Magistral Medium · reasoning04~6,677$0.01
Magistral Medium · no-reason013~8,675$0.02
Magistral Small · reasoning012~6,466$0.01
Magistral Small · no-reason013~6,342$0.01
Ministral 14B08~11,822$0.02
Ministral 8B08~10,571$0.02
Ministral 3B033~11,676$0.02
Mistral Nemo037~1,708$0.00

Token counts are estimated from artifact file size (chars ÷ 4). Cost = estimated output tokens × the model's published per-1M output price; input and reasoning tokens aren't counted, so true cost is higher — especially at high thinking effort. Generation time is wall-clock and only recorded for runs generated after 2026-07-02; it includes queue and throttling waits. Votes come from the public community tally. This page compares like-for-like one-shot generations, not lab benchmarks.