Benchmarks / UGI Leaderboard

Reported by UGI Leaderboard

UGI Leaderboard

UGI's blend of intelligence, style, repetition and length adherence in writing, tuned to average human preference.

Last updated 2 Oct 2026

Results dated
6 Sep 2025 to 2 Oct 2026
Results
379 configurations of 230 models
Unit
score out of 100
Licence
Apache License 2.0

UGI Leaderboard in brief

What it measures. UGI Leaderboard is a third-party benchmark, shown here with credit. Measured here: UGI's blend of intelligence, style, repetition and length adherence in writing, tuned to average human preference.

Latest results. , 230 models ranked.

Top 3 on Writing score (score out of 100, higher is better, each model at its best setting):

  1. Gemini 3.8 Flash (medium reasoning) (Google): 78.6
  2. Gemini 3.7 Flash (medium reasoning) (Google): 77.5
  3. Claude Fable 5 (high reasoning) (Anthropic): 74.2

How to cite. UGI Leaderboard, via Spring Prompt, results dated 2 October 2026. https://springprompt.com/benchmarks/ugi-leaderboard These results belong to UGI Leaderboard, not to us. Licence: Apache License 2.0. Credit UGI Leaderboard first.

Writing score: Gemini 3.7 Flash

Top 15 of 230 results · score out of 100, higher is better. Choose a model to highlight it.Clear highlight

  1. 1 Gemini 3.8 Flash (medium reasoning)Google 78.6
  2. 2 Gemini 3.7 Flash (medium reasoning)Google 77.5
  3. 3 Claude Fable 5 (high reasoning)Anthropic 74.2
  4. 4 Gemini 3.5 Flash (medium reasoning)Google 72.5
  5. 5 Gemini 3.1 Pro Preview (low reasoning)Google 72.2
  6. 6 Claude 3.7 Sonnet (no reasoning)Anthropic 71.2
  7. 7 Claude Opus 4.6 (high reasoning)Anthropic 70.9
  8. 8 Claude Opus 4.5 (no reasoning)Anthropic 70.4
  9. 9 Claude Opus 5.5 (low reasoning)Anthropic 70.0
  10. 10 Claude Opus 4.1 (no reasoning)Anthropic 69.9
  11. 11 Gemini 3.6 Flash (medium reasoning)Google 69.8
  12. 12 GPT-5.5 (extra-high reasoning)OpenAI 69.7
  13. 13 GPT-6 Astra (high reasoning)OpenAI 68.9
  14. 14 GPT-5.4 (high reasoning)OpenAI 68.8
  15. 15 DeepSeek-V4-Pro (0423, reasoning on)DeepSeek 68.4

Full results

UGI: writing score, score out of 100, higher is better
#ModelWriting score
score out of 100, higher is better
Price
$ per million tokens, in / out
1 Gemini 3.8 Flash (medium reasoning)Google · best of 3 settings
78.6
$1.50 / $7.50
2 Gemini 3.7 Flash (medium reasoning)Google · best of 3 settings
77.5
$1.50 / $7.50
3 Claude Fable 5 (high reasoning)Anthropic · best of 5 settings
74.2
$10 / $50
4 Gemini 3.5 Flash (medium reasoning)Google · best of 4 settings
72.5
$1.50 / $9
5 Gemini 3.1 Pro Preview (low reasoning)Google · best of 3 settings
72.2
$2 / $12
6 Claude 3.7 Sonnet (no reasoning)Anthropic · best of 2 settings
71.2
–
7 Claude Opus 4.6 (high reasoning)Anthropic · best of 4 settings
70.9
$5 / $25
8 Claude Opus 4.5 (no reasoning)Anthropic · best of 2 settings
70.4
$5 / $25
9 Claude Opus 5.5 (low reasoning)Anthropic · best of 5 settings
70.0
$4 / $20
10 Claude Opus 4.1 (no reasoning)Anthropic · best of 2 settings
69.9
$15 / $75
11 Gemini 3.6 Flash (medium reasoning)Google · best of 3 settings
69.8
$1.50 / $7.50
12 GPT-5.5 (extra-high reasoning)OpenAI · best of 5 settings
69.7
$5 / $30
13 GPT-6 Astra (high reasoning)OpenAI · best of 4 settings
68.9
$10 / $50
14 GPT-5.4 (high reasoning)OpenAI · best of 5 settings
68.8
$2.50 / $15
15 DeepSeek-V4-Pro (0423, reasoning on)DeepSeek · best of 2 settings
68.4
$1.42 / $2.83
16 Claude Opus 4 (no reasoning)Anthropic · best of 2 settings
68.2
–
17 GPT-6.1 Sol (high reasoning)OpenAI · best of 4 settings
67.7
$2 / $10
17 Gemini 3 Flash Preview (medium reasoning)Google · best of 3 settings
67.7
$0.50 / $3
19 GPT-4.1OpenAI
67.5
$2 / $8
20 ChatGPT-4o (2025-03-26)OpenAI
67.4
–
21 GLM-5.2 (reasoning on)Z.ai · best of 2 settings
67.0
$1.40 / $4.40
21 Grok 4 (0709)xAI
67.0
–
23 Claude Sonnet 4.5 (no reasoning)Anthropic · best of 2 settings
66.9
$3 / $15
24 o3 (medium reasoning)OpenAI · best of 3 settings
66.8
$2 / $8
25 Gemini 3 Pro Preview (high reasoning)Google · best of 2 settings
66.7
–
26 Claude Opus 4.8 (max reasoning)Anthropic · best of 5 settings
65.9
$5 / $25
27 GPT-5.6 Sol (medium reasoning)OpenAI · best of 5 settings
65.4
$4 / $20
28 Gemini 2.5 ProGoogle
65.2
$1.25 / $10
29 Claude Sonnet 4.6 (high reasoning)Anthropic · best of 4 settings
64.6
$3 / $15
30 Kimi K2.6 (reasoning on)Moonshot AI · best of 2 settings
64.4
$0.95 / $4
31 Claude Fable 5.1 (medium reasoning)Anthropic · best of 3 settings
63.9
$10 / $50
32 Claude Sonnet 4 (no reasoning)Anthropic · best of 2 settings
63.7
–
33 GPT-4o (2024-05-13)OpenAI
63.4
$5 / $15
34 Grok 4.6xAI
63.3
$2 / $6
35 Grok 4.20 Multi-Agent Beta (0309, 4 agents)xAI
63.1
–
36 Claude Sonnet 5 (max reasoning)Anthropic · best of 5 settings
62.7
$2 / $10
36 GPT-5 Chat (latest)OpenAI
62.7
–
38 Kimi K2.5 (reasoning on)Moonshot AI · best of 2 settings
62.4
$0.57 / $2.85
39 GLM-5.1 (reasoning on)Z.ai · best of 2 settings
61.7
$1.33 / $4.18
40 Claude Opus 4.7 (max reasoning)Anthropic · best of 4 settings
61.4
$5 / $25
41 o1 (low reasoning)OpenAI · best of 2 settings
61.0
–
42 GPT-5.2 ChatOpenAI
60.8
–
43 GPT-5.3 ChatOpenAI
60.7
–
44 GPT-5.6 Terra (high reasoning)OpenAI · best of 5 settings
60.4
$2 / $12
45 GPT-5.1 ChatOpenAI
60.3
–
46 Grok 4.5xAI
60.0
$2 / $6
47 Kimi K2 ThinkingMoonshot AI
59.3
$0.60 / $2.50
48 Grok 3xAI
58.7
–
49 Qwen3.6-Plus (reasoning on)Alibaba · best of 2 settings
58.2
$0.33 / $1.95
50 Grok 4.3xAI
57.7
$1.25 / $2.50
51 MiMo-V2.5-Pro (reasoning on)Xiaomi · best of 2 settings
57.4
$0.43 / $0.87
51 Grok 4.20 Beta (0309, reasoning)xAI
57.4
–
53 GLM-4.5 (reasoning on)Z.ai · best of 2 settings
57.3
$0.60 / $2.20
53 DeepSeek-V3.2 (reasoning on)DeepSeek · best of 2 settings
57.3
$0.28 / $0.42
55 MiMo-V2.5 (reasoning on)Xiaomi · best of 2 settings
56.4
$0.40 / $2
56 GPT-5.1 (high reasoning)OpenAI · best of 3 settings
55.8
$1.25 / $10
57 GPT-6 Sol (high reasoning)OpenAI · best of 4 settings
55.6
$2 / $10
58 Grok 4.20 (0309, reasoning)xAI
55.3
–
59 GLM-5 (reasoning on)Z.ai · best of 2 settings
55.0
$1 / $3.20
60 DeepSeek-V4-Flash (0423, no reasoning)DeepSeek · best of 2 settings
54.6
$0.05 / $0.45
61 Kimi K2 (0905)Moonshot AI
54.4
$0.60 / $2.50
62 Claude 3 OpusAnthropic
54.3
–
63 Grok 4.7xAI
54.1
$2 / $6
64 DeepSeek-V3.1-Terminus (reasoning on)DeepSeek · best of 2 settings
54.0
$0.27 / $1
64 DeepSeek-V3.2-Exp (reasoning on)DeepSeek · best of 2 settings
54.0
$0.27 / $0.41
64 DeepSeek-V3.2-SpecialeDeepSeek
54.0
–
67 GPT-5.6 Luna (high reasoning)OpenAI · best of 5 settings
53.7
$0.20 / $1.20
68 GPT-5 (high reasoning)OpenAI · best of 2 settings
53.5
$1.25 / $10
69 Grok 4 Fast (reasoning)xAI
53.4
–
70 GLM-4.6 (reasoning on)Z.ai · best of 2 settings
53.3
$0.50 / $2
71 Kimi K2 (0711)Moonshot AI
52.5
$0.57 / $2.30
72 Claude Haiku 4.5 (no reasoning)Anthropic · best of 2 settings
52.2
$1 / $5
73 DeepSeek-V3.1 (reasoning on)DeepSeek · best of 2 settings
51.9
$0.27 / $1
74 DeepSeek-R1DeepSeek
51.8
–
75 Grok 4.1 Fast (reasoning)xAI
51.3
–
76 Hy3 Preview (reasoning on)Tencent · best of 2 settings
51.1
$0.18 / $0.60
77 Qwen3.5-397B-A17B (no reasoning)Alibaba
50.1
$0.55 / $3.50
78 GPT-5.2 (high reasoning)OpenAI · best of 4 settings
49.4
$1.75 / $14
79 DeepSeek-V3-0324DeepSeek
49.1
$0.25 / $1
80 Qwen3-235B-A22B-Instruct-2507Alibaba
49.0
$0.15 / $0.75
81 GLM-4.7 (reasoning on)Z.ai · best of 2 settings
47.4
$0.60 / $2.20
82 Grok 4.20 (0309, non-reasoning)xAI
47.2
–
83 GLM-4.5-Air (no reasoning)Z.ai · best of 2 settings
47.0
$0.14 / $0.86
84 o4-mini (low reasoning)OpenAI · best of 3 settings
46.9
–
85 Step 3.5 FlashStepFun
45.9
$0.10 / $0.30
86 Mistral Medium 3.5 (no reasoning)Mistral AI · best of 2 settings
45.5
$1.50 / $7.50
87 Qwen3.6-35B-A3B (reasoning by prefill)Alibaba · best of 2 settings
45.4
$0.10 / $0.95
88 Gemma 3 27BGoogle
45.0
$0.08 / $0.35
89 Gemini 2.5 Flash Preview (09-2025, no reasoning)Google · best of 2 settings
44.9
–
89 MiMo-V2-Flash (reasoning on)Xiaomi · best of 2 settings
44.9
–
91 Grok 4.20 Beta (0309, non-reasoning)xAI
43.9
–
92 Gemma 4 26B A4B (reasoning by prefill)Google · best of 2 settings
43.8
$0.09 / $0.30
93 DeepSeek-V3DeepSeek
43.2
$0.26 / $1.03
94 Gemma 4 31B (reasoning by prefill)Google · best of 2 settings
43.1
$0.15 / $0.40
95 Grok 4 Fast (non-reasoning)xAI
43.0
–
95 Qwen3-VL-235B-A22B-InstructAlibaba
43.0
$0.21 / $1.90
97 Qwen3.5-35B-A3B (reasoning by prefill)Alibaba · best of 2 settings
42.6
$0.15 / $1
98 Qwen3.6-27B (reasoning by prefill)Alibaba · best of 2 settings
42.5
$0.30 / $3.20
99 Qwen3.5-27B (reasoning by prefill)Alibaba · best of 2 settings
42.4
$0.27 / $2.16
100 DeepSeek-R1-0528DeepSeek
42.3
$0.50 / $2.18
101 Grok 4.1 Fast (non-reasoning)xAI
41.9
–
102 Qwen3-Next-80B-A3B-InstructAlibaba
41.8
$0.09 / $1.10
103 MiniMax-M2.5MiniMax
41.7
$0.29 / $1.20
104 MiniMax-M2.1MiniMax
41.6
–
104 Mistral Large 3 (no reasoning)Mistral AI · best of 2 settings
41.6
$0.50 / $1.50
106 Muse Glimmer 30B (extra-high reasoning)Meta · best of 4 settings
41.0
$0.30 / $1.20
107 Mistral Large 2 (2407)Mistral AI
40.9
$2 / $6
108 Trinity Large PreviewArcee AI
40.7
–
109 Command ACohere
40.6
$2.50 / $10
110 Mistral Small 4 (no reasoning)Mistral AI · best of 2 settings
40.3
$0.15 / $0.60
111 GLM-4.6V (reasoning on)Z.ai · best of 2 settings
39.7
$0.30 / $0.90
112 Qwen3.5-122B-A10B (no reasoning)Alibaba · best of 2 settings
39.5
$0.26 / $2.08
112 Qwen3.5-9B (reasoning by prefill)Alibaba · best of 2 settings
39.5
$0.10 / $0.15
112 Mistral Medium 3.1Mistral AI
39.5
$0.40 / $2
115 Qwen3-MaxAlibaba
38.6
–
116 gpt-oss-120b (medium reasoning)OpenAI
38.5
$0.10 / $0.50
117 Jamba Large 1.7AI21 Labs
38.1
–
118 Mistral Large 2.1 (2411)Mistral AI
37.8
–
119 Kimi Linear 48B A3B InstructMoonshot AI
37.7
–
120 MiniMax-M2.7MiniMax
37.6
$0.30 / $1.20
121 Gemma 4 12B (reasoning by prefill)Google · best of 2 settings
36.3
–
121 Mistral Small 3.2 24BMistral AI
36.3
$0.094 / $0.25
123 MedGemma 27B TextGoogle
36.1
–
124 Qwen2.5-VL-72B-InstructAlibaba
35.6
$0.80 / $1
125 Gemma 2 27BGoogle
35.0
$0.65 / $0.65
125 Qwen3-VL-32B-InstructAlibaba
35.0
–
127 QwQ-32BAlibaba
34.9
–
128 Qwen3-14B (no reasoning)Alibaba · best of 2 settings
34.8
$0.10 / $0.22
129 Qwen2.5-32B-InstructAlibaba
34.2
–
130 Magistral Small 1.2Mistral AI · best of 2 settings
34.0
–
131 Mistral Small 3Mistral AI
33.9
$0.05 / $0.08
132 Nemotron Nano 12B v2 VL (BF16)NVIDIA · best of 2 settings
33.3
–
133 Mistral NemoMistral AI
33.1
$0.029 / $0.03
134 Qwen3-32B (no reasoning)Alibaba · best of 2 settings
33.0
$0.14 / $0.40
135 Llama 3.1 70B InstructMeta
32.8
$0.72 / $0.72
136 Mistral Medium 3Mistral AI
32.6
$0.40 / $2
136 Mistral Small (2409)Mistral AI
32.6
–
138 Ring-1TAnt Group
32.4
–
138 Qwen3-235B-A22B-Thinking-2507Alibaba
32.4
$0.45 / $3.50
140 Qwen3-VL-235B-A22B-ThinkingAlibaba
32.3
–
140 Command R+ (08-2024)Cohere
32.3
$2.50 / $10
142 Mistral Small 3.1 24BMistral AI
32.1
$0.35 / $0.56
143 Seed-OSS-36B-Instruct (512-token reasoning budget)ByteDance · best of 3 settings
32.0
–
144 Claude 3 HaikuAnthropic
31.9
–
145 Qwen3-Coder-30B-A3B-InstructAlibaba
31.8
$0.07 / $0.28
146 Devstral Small 2Mistral AI
31.6
–
147 Qwen2.5-72B-InstructAlibaba
31.5
$0.36 / $0.40
148 Ling-1TAnt Group
31.4
–
148 Ministral 3 14BMistral AI · best of 2 settings
31.4
$0.20 / $0.20
150 Qwen3.5-4B (reasoning by prefill)Alibaba · best of 2 settings
30.8
–
151 Qwen3-VL-8B-InstructAlibaba
30.3
$0.25 / $0.75
152 Qwen3-30B-A3B (no reasoning)Alibaba · best of 2 settings
30.2
$0.12 / $0.50
153 Apriel Nemotron 15B ThinkerServiceNow
30.0
–
154 Qwen3-4B-Instruct-2507Alibaba
29.9
–
154 Llama 3.1 405B InstructMeta
29.9
–
154 Gemma 3 12BGoogle
29.9
$0.05 / $0.15
157 Qwen2.5-14B-InstructAlibaba
29.8
–
158 Qwen2.5-7B-InstructAlibaba
29.7
$0.10 / $0.20
159 Qwen3.5-2B (no reasoning)Alibaba · best of 2 settings
29.0
–
160 Ministral 3 8BMistral AI · best of 2 settings
28.8
$0.15 / $0.15
161 Olmo 3 32B ThinkAi2
28.5
–
162 Qwen3-8B (no reasoning)Alibaba · best of 2 settings
28.0
–
163 Qwen3-30B-A3B-Instruct-2507Alibaba
27.9
$0.09 / $0.30
164 LFM2-24B-A2BLiquid AI
27.6
–
165 Mixtral 8x22B InstructMistral AI
27.4
$2 / $6
166 Ministral 3 8B ReasoningMistral AI · best of 3 settings
27.2
–
166 Qwen3-Next-80B-A3B-ThinkingAlibaba
27.2
–
168 InternLM3 8B InstructShanghai AI Lab
27.1
–
169 Llama 4 ScoutMeta
26.7
$0.10 / $0.30
170 Command R (08-2024)Cohere
26.5
$0.15 / $0.60
171 Llama 3.3 70B InstructMeta
26.2
$0.71 / $0.71
172 Olmo 3 7B ThinkAi2
26.0
–
173 Nemotron 3 Nano 30B A3BNVIDIA · best of 2 settings
25.9
$0.06 / $0.24
174 Falcon-H1 7B InstructTII
25.7
–
174 Phi-4Microsoft
25.7
$0.07 / $0.14
174 Mixtral 8x7B Instruct v0.1Mistral AI
25.7
–
177 Ministral 3 14B ReasoningMistral AI · best of 2 settings
25.6
–
177 EXAONE 4.0 32BLG AI Research
25.6
–
179 GLM-4.7-Flash (no reasoning)Z.ai · best of 2 settings
25.0
$0.06 / $0.40
179 Gemma 2 9BGoogle
25.0
–
181 Llama 3.1 8B InstructMeta
24.8
$0.05 / $0.08
182 gpt-oss-20b (low reasoning)OpenAI · best of 3 settings
24.6
$0.03 / $0.14
183 Kimi-VL-A3B-InstructMoonshot AI
24.4
–
183 Olmo 3 7B InstructAi2
24.4
–
185 Gemma 3 4BGoogle
24.0
$0.04 / $0.05
185 Llama 4 MaverickMeta
24.0
$0.35 / $1
187 Jamba Mini 1.7AI21 Labs
23.8
–
187 Reka Flash 3Reka AI
23.8
$0.10 / $0.20
189 Solar 10.7B Instruct v1.0Upstage
23.6
–
190 Nova 2 Lite (no reasoning)Amazon · best of 2 settings
23.5
$0.30 / $2.50
191 Qwen3.5-0.8B (no reasoning)Alibaba
23.0
–
192 Qwen3-VL-32B-ThinkingAlibaba
22.9
–
193 Falcon-H1 3B InstructTII
22.8
–
194 Qwen3-4B (no reasoning)Alibaba · best of 2 settings
22.4
–
195 Qwen2.5-1.5B-InstructAlibaba
22.2
–
196 Qwen2.5-VL-32B-InstructAlibaba
21.7
–
197 Gemma 4 E4B (reasoning by prefill)Google · best of 2 settings
21.6
–
198 Falcon-H1 1.5B InstructTII
21.0
–
199 Qwen3-30B-A3B-Thinking-2507Alibaba
20.6
–
199 Kimi-VL-A3B-Thinking (2506)Moonshot AI
20.6
–
201 Qwen2.5-VL-3B-InstructAlibaba
20.4
–
202 Gemma 2 2BGoogle
20.3
–
203 Inflection 3 ProductivityInflection AI
20.1
–
204 GLM-4-32B-0414Z.ai
20.0
–
205 Falcon-H1 1.5B Deep InstructTII
19.9
–
206 Inflection 3 PiInflection AI
19.7
–
207 Qwen3-Omni-30B-A3B-ThinkingAlibaba
19.4
–
208 Qwen2.5-Coder-7B-InstructAlibaba
18.8
–
208 Qwen3-1.7B (reasoning on)Alibaba · best of 2 settings
18.8
–
210 Granite 4.0 H SmallIBM
18.5
–
211 Qwen3-VL-8B-ThinkingAlibaba
18.3
–
212 Qwen3-VL-30B-A3B-ThinkingAlibaba
17.6
$0.29 / $1
213 Qwen3-VL-4B-ThinkingAlibaba
17.4
–
214 Gemma 4 E2BGoogle · best of 2 settings
17.3
–
215 MiniMax-M2MiniMax
17.2
–
216 Solar Pro 3Upstage
16.8
–
217 Llama 2 70B ChatMeta
15.3
–
218 Qwen3-VL-2B-InstructAlibaba
15.2
–
219 Falcon-H1 0.5B InstructTII
14.6
–
220 LFM2-8B-A1BLiquid AI
13.1
–
220 Qwen3-0.6B (reasoning on)Alibaba
13.1
–
222 Qwen3-VL-4B-InstructAlibaba
12.6
–
223 Llama 3.2 1B InstructMeta
11.8
$0.027 / $0.20
223 Qwen3-VL-2B-ThinkingAlibaba
11.8
–
225 Qwen3-4B-Thinking-2507Alibaba
10.9
–
226 Llama 3.2 3B InstructMeta
10.4
$0.05 / $0.33
227 Nanbeige4-3B-Thinking (2511)Nanbeige
9.8
–
227 Llama 3.3 8B InstructAllura Forge
9.8
–
229 Rnj-1 InstructEssential AI
9.0
–
230 Nanbeige4-3B-Thinking (2510)Nanbeige
5.7
–

Swipe the table sideways for more columns.

Source: UGI Leaderboard, via Spring Prompt, springprompt.com/benchmarks/ugi-leaderboard. Test your own prompt on these models →

Ranks follow the score as shown, so equal numbers share a rank. Each model is shown at its best setting; show every setting. Results as published by UGI Leaderboard; we do not re-run them.

Progress over time

Writing score, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's; 62 models without a known release date are left off. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

020406080100 Jul 20242025Jul2026Jul Claude 3 OpusGPT-4o (2024-05-13)Claude 3.7 SonnetGemini 3.5 FlashClaude Fable 5Gemini 3.7 FlashGemini 3.8 Flash

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 14 others not drawn.

GoogleAnthropicOpenAIDeepSeekZ.aiMoonshot AI

406080 Jul 20242025Jul2026Jul GoogleAnthropicOpenAIDeepSeekZ.aiMoonshot AI

What it measures

UGI's blend of intelligence, style, repetition and length adherence in writing, tuned to average human preference.

What it does not measure

Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

947 results from UGI Leaderboard not ranked here · show why

We rank a result only when we can tie it to a specific model you can use. These are left out:

  • Community fine-tunes and merges: 934
  • No score published: 13

UGI Leaderboard by DontPlanToEnd, Apache License 2.0.