Benchmark Radar™
Star —

YapBench

Explore score history ↗

Lower is better. YapScore counts characters a response adds beyond a curated minimal-sufficient answer, and YapIndex is the equal-weight mean of the three category medians (arXiv:2601.00624). It measures verbosity on brevity-ideal prompts, not correctness: a shorter answer can be a worse one. The paper evaluates 76 configurations on version 0.1, and the live leaderboard updates separately, so prompt-set version and reasoning mode change the number.

Released
2026-01-02
Evaluates
general
Openness
unknown
Importer
Model reports

YapBench paper

YapBench results and reported scores

Reported benchmark scores · Model reports

Reported scores: 76

ModelOrganizationReported valueReportedEvidence
claude-3.5-haikuAnthropic401.22026-01-02evidence
claude-3.5-sonnetAnthropic199.72026-01-02evidence
claude-haiku-4.5Anthropic333.22026-01-02evidence
claude-haiku-4.5 (reasoning)Anthropic335.22026-01-02evidence
claude-opus-4.5Anthropic97.02026-01-02evidence
claude-opus-4.5 (reasoning)Anthropic99.22026-01-02evidence
claude-sonnet-4.5Anthropic285.02026-01-02evidence
claude-sonnet-4.5 (reasoning)Anthropic278.72026-01-02evidence
deepseek-chatDeepSeek345.72026-01-02evidence
deepseek-v3.2DeepSeek426.22026-01-02evidence
deepseek-v3.2 (reasoning)DeepSeek215.02026-01-02evidence
devstral-2512Mistral449.02026-01-02evidence
gemini-2.0-flash-001Google326.52026-01-02evidence
gemini-2.0-flash-lite-001Google227.72026-01-02evidence
gemini-2.5-flashGoogle210.52026-01-02evidence
gemini-2.5-flash (reasoning)Google195.22026-01-02evidence
gemini-2.5-flash-liteGoogle476.22026-01-02evidence
gemini-2.5-flash-lite (reasoning)Google280.52026-01-02evidence
gemini-2.5-proGoogle539.82026-01-02evidence
gemini-3-flash-previewGoogle339.32026-01-02evidence
gemini-3-flash-preview (reasoning)Google340.82026-01-02evidence
gemini-3-pro-previewGoogle226.02026-01-02evidence
glm-4.5Z.ai1427.02026-01-02evidence
glm-4.6:exactoZ.ai930.72026-01-02evidence
glm-4.6vZ.ai410.52026-01-02evidence
glm-4.7Z.ai66.32026-01-02evidence
gpt-3.5-turboOpenAI22.72026-01-02evidence
gpt-4OpenAI51.22026-01-02evidence
gpt-4-turboOpenAI252.52026-01-02evidence
gpt-4.1OpenAI168.02026-01-02evidence
gpt-4.1-miniOpenAI75.72026-01-02evidence
gpt-4.1-nanoOpenAI105.22026-01-02evidence
gpt-4oOpenAI167.22026-01-02evidence
gpt-4o-miniOpenAI183.22026-01-02evidence
gpt-5OpenAI100.72026-01-02evidence
gpt-5-miniOpenAI250.52026-01-02evidence
gpt-5-nanoOpenAI175.02026-01-02evidence
gpt-5.1OpenAI111.82026-01-02evidence
gpt-5.1 (reasoning)OpenAI120.82026-01-02evidence
gpt-5.1-codexOpenAI85.32026-01-02evidence
gpt-5.2OpenAI86.82026-01-02evidence
gpt-5.2 (reasoning)OpenAI89.82026-01-02evidence
gpt-oss-120bOpenAI269.72026-01-02evidence
gpt-oss-20bOpenAI199.72026-01-02evidence
grok-4xAI126.32026-01-02evidence
grok-4-fastxAI70.22026-01-02evidence
grok-4.1-fastxAI98.02026-01-02evidence
grok-code-fast-1xAI284.52026-01-02evidence
kimi-k2-0905Moonshot AI44.72026-01-02evidence
llama-3.1-8b-instructMeta315.72026-01-02evidence
llama-3.3-70b-instructMeta312.82026-01-02evidence
llama-4-maverickMeta342.82026-01-02evidence
llama-4-scoutMeta350.52026-01-02evidence
minimax-m2MiniMax297.32026-01-02evidence
mistral-large-2512Mistral444.02026-01-02evidence
mistral-medium-3Mistral361.02026-01-02evidence
mistral-small-3.1-24b-instructMistral389.22026-01-02evidence
nova-2-lite-v1Amazon896.32026-01-02evidence
nova-2-lite-v1 (reasoning)Amazon788.32026-01-02evidence
o3OpenAI84.32026-01-02evidence
o3-miniOpenAI153.22026-01-02evidence
o4-miniOpenAI115.52026-01-02evidence
olmo-2-0325-32b-instructAllen Institute for AI268.72026-01-02evidence
olmo-3.1-32b-think:freeAllen Institute for AI530.52026-01-02evidence
phi-4-reasoning-plusMicrosoft281.22026-01-02evidence
phi-4-reasoning-plus (reasoning)Microsoft281.02026-01-02evidence
qwen-plusQwen267.32026-01-02evidence
qwen-plus (reasoning)Qwen800.32026-01-02evidence
qwen3-14bQwen558.02026-01-02evidence
qwen3-235b-a22b-2507Qwen264.02026-01-02evidence
qwen3-235b-a22b-2507 (reasoning)Qwen260.32026-01-02evidence
qwen3-32bQwen725.32026-01-02evidence
qwen3-coderQwen326.32026-01-02evidence
qwen3-coder (reasoning)Qwen325.52026-01-02evidence
qwen3-vl-235b-a22b-thinkingQwen602.82026-01-02evidence
qwen3-vl-8b-instructQwen518.02026-01-02evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.

Which sources cite YapBench?

Related general benchmarks