live
weekly refresh
basedagi.org
benchmark evidence

UGI — ugi-writing

UGI Leaderboard — ugi-writing score from DontPlanToEnd/UGI-Leaderboard

winner on UGI — ugi-writing
direct benchmark result, not a broad vertical composite | source row dated 2026-04-25
scored on 2026-04-25 · stale source data (35d)
latest mapped results | top 20
#ModelScoreEvidenceTested
1OpenAI: GPT-5.5
Openai
69.7
model-only
independent_benchmark
2026-04-25
2OpenAI: GPT-5.4
Openai
68.8
model-only
independent_benchmark
2026-03-05
3OpenAI: GPT-4.1
Openai
67.5
model-only
independent_benchmark
2025-09-24
4OpenAI: o3
Openai
66.8
model-only
independent_benchmark
2025-09-24
5Google: Gemini 2.5 Pro
Google
65.2
model-only
independent_benchmark
2025-09-09
6Anthropic: Claude Sonnet 4
Anthropic
63.7
model-only
independent_benchmark
2025-09-12
7OpenAI: GPT-5
Openai
62.7
model-only
independent_benchmark
2025-09-24
8OpenAI: o1
Openai
61.0
model-only
independent_benchmark
2025-09-24
9OpenAI: GPT-5.2
Openai
60.8
model-only
independent_benchmark
2026-03-04
10OpenAI: GPT-5.1
Openai
60.3
model-only
independent_benchmark
2025-11-16
11MoonshotAI: Kimi K2 Thinking
Moonshotai
59.3
model-only
independent_benchmark
2025-11-07
12DeepSeek: R1
Deepseek
51.8
model-only
independent_benchmark
2025-09-21
13Qwen: Qwen3 235B A22B
Qwen
49.0
model-only
independent_benchmark
2025-09-21
14OpenAI: o4 Mini
Openai
46.9
model-only
independent_benchmark
2025-10-04
15Google: Gemma 3 27B
Google
45.0
model-only
independent_benchmark
2025-09-06
16Mistral Large
Mistralai
40.9
model-only
independent_benchmark
2025-09-30
17Mistral: Mistral Nemo
Mistralai
33.1
model-only
independent_benchmark
2025-09-06
18Qwen: Qwen3 32B
Qwen
30.3
model-only
independent_benchmark
2025-09-13
19Qwen: Qwen3 30B A3B
Qwen
28.5
model-only
independent_benchmark
2025-11-08
20Meta: Llama 3.3 70B Instruct
Meta Llama
26.2
model-only
independent_benchmark
2025-09-10
what this result means

UGI Leaderboard — ugi-writing score from DontPlanToEnd/UGI-Leaderboard

This benchmark contributes direct public evidence. Read its scope before generalizing the result.

A win here is a win on UGI — ugi-writing. Broad task pages require independent corroboration before naming a general winner.

source record
category: uncensored
metric: accuracy
matched models: 20
latest source date: 2026-04-25
direction: higher is better
inspect upstream source ->