Dataclaw Vaynelee: 21 Claude Code Sessions with 3.4M Input Tokens

Aggregating 21 Claude Code conversation sessions involving 3.4M input tokens and 5K output tokens, updated by vaynelee in February 2026. The records capture agentic coding interactions using the glm-4.7 model across two distinct projects. It provides a specialized look at high-context tool-use and coding assistant behavior.

Use Cases

• Analyzing agentic coding patterns using the git_branch field
• Benchmarking glm-4.7 performance in high-context coding tasks
• Studying tool-use behavior in Claude Code sessions

Strengths

• Contains 3.4M input tokens for context-heavy analysis
• Includes git_branch metadata for version control context
• MIT licensed for open redistribution

Limitations

• Small sample size of 21 sessions
• Extreme imbalance between input (3.4M) and output (5K) tokens
• Restricted to a single model type (glm-4.7)

Provenance

Source

vaynelee via DataClaw export

Collection Method

Exported logs from Claude Code sessions

Time Range

Ending February 2026

Freshness

Last updated February 27, 2026.