live
weekly refresh
basedagi.org
benchmark evidence

LiveBench Instruction Following

LiveBench instruction_following benchmark.

winner on LiveBench Instruction Following
direct benchmark result, not a broad vertical composite | source row dated 2026-05-20
scored on 2026-05-20
latest mapped results | top 20
#ModelScoreEvidenceTested
1DeepSeek: DeepSeek V3.2
Deepseek
86.8
model-only
independent_benchmark
2026-05-20
2Google: Gemini 2.0 Flash Lite
Google
84.2
model-only
independent_benchmark
2026-05-20
3Google: Gemini 2.5 Pro
Google
83.2
model-only
independent_benchmark
2026-05-20
4OpenAI: o1
Openai
83.2
model-only
independent_benchmark
2026-05-20
5DeepSeek: R1
Deepseek
80.6
model-only
independent_benchmark
2026-05-20
6OpenAI: o3 Mini
Openai
79.8
model-only
independent_benchmark
2026-05-20
7Meta: Llama 4 Maverick
Meta Llama
79.6
model-only
independent_benchmark
2026-05-20
8Meta: Llama 3.3 70B Instruct
Meta Llama
79.0
model-only
independent_benchmark
2026-05-20
9Google: Gemini 2.0 Flash
Google
75.9
model-only
independent_benchmark
2026-05-20
10Google: Gemma 3 27B
Google
71.5
model-only
independent_benchmark
2026-05-20
11Amazon: Nova Pro 1.0
Amazon
70.1
model-only
independent_benchmark
2026-05-20
12OpenAI: GPT-5
Openai
70.1
model-only
independent_benchmark
2026-05-20
13Mistral Large
Mistralai
69.9
model-only
independent_benchmark
2026-05-20
14DeepSeek: DeepSeek V4 Pro
Deepseek
65.2
model-only
independent_benchmark
2026-05-08
15Qwen2.5 Coder 32B Instruct
Qwen
64.9
model-only
independent_benchmark
2026-05-20
16Mistral: Mistral Small 4
Mistralai
62.4
model-only
independent_benchmark
2026-05-20
what this result means

LiveBench instruction_following benchmark.

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

A win here is a win on LiveBench Instruction Following. Broad task pages require independent corroboration before naming a general winner.

source record
category: overall
metric: accuracy
matched models: 16
latest source date: 2026-05-20
direction: higher is better
inspect upstream source ->