Tachibana4: Agentic Coding Tasks for DeepSeek-V4-Pro

Sequelbox's Tachibana4 dataset tests agentic coding skills with real-world, challenging tasks across multiple programming languages. Synthetic prompts utilize varied personas, experience levels, and communication styles to maximize real-world flexibility. The dataset was last updated on May 7, 2026.

Use Cases

• Benchmarking code generation models based on real-world, challenging agentic tasks
• Training models for multi-language code generation based on diverse programming topics
• Evaluating model adaptability based on prompts with varied personas and communication styles
• Assessing code generation robustness based on tasks designed for different experience levels

Strengths

• Focuses on real-world, challenging agentic coding tasks
• Prompts utilize a variety of personas, experience levels, and communication styles
• Covers a variety of programming languages and topics

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

sequelbox

Collection Method

Likely contains synthetically generated prompts and tasks.

Freshness

Last updated 2026-05-07 01:09:32; freshness should be verified