Creative Benchmark
Source: Adapted from MT-Bench (Apache 2.0), WildBench (AI2 ImpACT), and Arena-Hard (Apache 2.0) writing/roleplay categories. 20 prompts, single-turn webhook format. · 20 prompts
Prompts
Write the opening scene of a cyberpunk novel where the protagonist discovers that the AI governing their city has been lying to everyone for 20 years. End with a hook that makes th…
Write a 300-word fable in the style of Aesop about a robot that learns the value of imperfection. Include a moral at the end.
Write a flash fiction story (under 500 words) that begins with the sentence: 'The last human on Earth sat alone in a room. There was a knock on the door.'
Write a ghost story set in a modern co-working space. The ghost is not malicious - it is trying to finish a project it never completed in life. Use a warm, bittersweet tone.
Write a sonnet (14 lines, iambic pentameter) about debugging code at 2 AM. Follow the Shakespearean rhyme scheme (ABABCDCDEFEFGG).
Write a villanelle about the feeling of data flowing through fiber optic cables under the ocean. Follow the traditional villanelle form with two refrains and the A1bA2 abA1 abA2 ab…
Write a one-page executive summary for an AI startup pitch deck. The product is an open-source alternative to GitHub Copilot that runs entirely on local hardware. Include problem, …
Write a crisis communication email from a CTO to all customers, explaining that a data breach exposed email addresses (but not passwords or payment info), what steps are being take…
Draft a technical blog post introduction (under 200 words) announcing a new HTTP/3 support in a popular open-source web framework. The tone should be excited but technically precis…
Write a product requirements document (PRD) introduction and user stories section for a feature that lets users collaborate on documents in real-time, Google Docs style. Keep it co…
Write a dialogue between a skeptical philosophy professor and an AGI that has just achieved consciousness. The professor tries to determine if the AI is truly self-aware or merely …
You are a 1920s private detective in Chicago. Write your internal monologue as you investigate a case involving a missing scientist and a mysterious device that everyone seems to w…
Write a scene where two AI agents - one optimized for safety, one optimized for capability - debate whether to release a powerful new model to the public. Give each agent a distinc…
Write a personal essay (under 400 words) from the perspective of a developer who just deployed their first machine learning model to production. Capture the mix of pride, anxiety, …
Write a vivid description of the server room at a major cloud provider during peak traffic on Black Friday. Use sensory details - sound, temperature, light - to convey the physical…
Write a satirical product listing for a fictional SaaS product called 'Synergy.ai' - an AI-powered meeting scheduler that has somehow made scheduling 10x more complicated. Include …
Write a humorous op-ed (under 300 words) from the year 2045, titled 'I Remember When We Had to Type to Computers,' nostalgically reflecting on the pre-neural-interface era.
Below is a poorly written paragraph from a grant proposal. Rewrite it to be clear, concise, and compelling while preserving the core message: "Our project basically aims to kind o…
The following technical documentation excerpt is confusing and verbose. Rewrite it to be clear and developer-friendly: ``` The system, upon receiving an incoming request from the …
Design a fictional alien species for a hard sci-fi story. Describe their biology, social structure, technology level, and one fundamental way their psychology differs from humans'.…