GATC

A final diff shows the result. A session shows how the agent got there.

The diff leaves out the commands that failed, the files the agent looked for and did not find, and the follow-up requests that changed the task. This dataset records them.

Every step, in the order it happened

Each session keeps the user’s messages, the agent’s replies, every tool call with its arguments and, where the agent recorded one, its result, the reasoning and token counts where the agent logged them, and each failed call together with what the agent did next.

How agents work, and where they fail

How it differs from other agent datasets

Every cell was checked twice against each dataset’s card, files or paper (October 2026). The Size column says what kind of session each dataset holds.

✓ yes, in most sessions · ◐ partly, or in a minority · ✗ no. Failures ◐: failed calls are labelled, with no recovery flag; ours is computed by us (a tool call within the next four steps succeeded). Code changes: the change itself (a diff, a patch, or an edit with its old and new text); AgentLogs keeps line counts only. Scanned before release: a scan of the released files with its result stated (◐: tools named, no result). Interactive: a person typed the prompts for their own work, by a blind review of 360 sessions by a language model; these sessions hold about two-thirds of all steps. Datasets of real sessions can overlap, GATC included. Datasets of agents run only on prepared tasks (such as NVIDIA’s Open-SWE-Traces) are not listed.

What the sessions do

Each session gets a task type from the tools it called: implementation if a write or edit tool was called; otherwise exploration if a read, grep, glob, list, web search or web fetch tool was called; otherwise conversation, if the agent replied (shell commands may still run).

Share of all 117,796 sessions, and the number of sessions. The other 2,305 sessions (2.0%) have no agent reply.

Which tools agents call, and how often each one fails

Tool names are normalised across agents, so a shell command is bash whatever the agent called it; task hands work to a sub-agent, and other holds tool names outside the normalised vocabulary. Of the common tools, fetching a web page fails most often; the shell fails on one call in eighteen.

The eleven most-called tools: share of all 5,751,020 tool calls, and the share of each tool’s calls that failed.

Sessions by the model tier they used

Each model is shown as a capability tier: S frontier flagship, A frontier, B workhorse, C small and fast, D legacy. A session counts under the highest tier it used.

102,381 sessions used a model with a tier. Token medians cover the sessions that report output tokens. Tiers differ in agent and task as well as in model, so the columns describe the data, not the models alone.

Five tables, joined on one key

Every table joins on trajectory_id, and a per-tool summary (tool_grammar) comes with them. Row counts are for the full dataset; the public sample has the same tables at 1% of the size, and its dataset viewer opens any session in the browser.

from datasets import load_dataset

# the public 1% sample; each table is a named config
messages = load_dataset("GOAT-AI/gatcv6", "messages", split="train")
errors = load_dataset("GOAT-AI/gatcv6", "error_recovery", split="train")
recovered = errors.filter(lambda r: r["recovered"])

Checked before release

People, accounts, organisations, e-mail addresses, phone numbers, network addresses and secrets are replaced by typed placeholders that stay the same within a session: <USER_1>, <ORG_2>, <KEYBODY_1>. Where a value could not be replaced safely, the text that held it was removed and marked <WITHHELD>. Names of third-party open-source projects and their owners, weak or default values in password fields, and names that are also ordinary words can remain; the residual risk is not zero.

The data is pseudonymised, not anonymous: names that remain in the text, such as package or directory names, and the code itself can still point to its author. To have a session removed, write to us with its trajectory_id.

Sample sessions on Hugging Face

Each of the 1,189 sample sessions opens in the Hugging Face dataset viewer.

Only the sessions recorded with frontier flagship models: GATC-S

Cite the dataset

@misc{goat_agentic_coding_sessions_2026,
  title        = {{GATC}: Agentic Coding Sessions},
  author       = {{GOAT labs}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/GOAT-AI/gatcv6}},
  note         = {Apache-2.0 for the 1\% sample}
}