XYZ-Aquila is a family of search agents produced by a bounded, verification-gated AI-for-AI loop: humans set the objective and acceptance gate, while AI agents optimize data, post-training, runtime, evaluation, and infrastructure inside that envelope.
XYZ-Aquila-mini and XYZ-Aquila-pro are evaluated on BrowseComp, BrowseComp-ZH, DeepSearchQA, GAIA, LiveBrowseComp, Humanity’s Last Exam, and WideSearch. The external benchmark suite is isolated from the optimization loop; development decisions are made through a private, answer-hidden gate.
| Model | BrowseComp | BrowseComp-ZH | DeepSearchQA | GAIA | LiveBrowseComp | HLE | WideSearch |
|---|---|---|---|---|---|---|---|
| Agents-A1 | 75.5 | — | — | 96.0 | 29.6† | 47.6 | — |
| Nex-N2-mini | 74.1 | 79.6† | 87.2† | — | 41.4† | 37.1† | 62.0 |
| apodex-mini | 71.5 | 80.6 | 82.2 | — | 32.8† | 46.8 | — |
| MiroThinker 1.7 mini | 67.9 | — | — | 80.3 | 34.9† | 36.4 | 73.3† |
| XYZ-Aquila-mini | 78.8 | 82.9 | 89.5 | 97.1 | 48.7 | 51.1 | 80.8 |
Percentages are reported from the slim technical report. DeepSearchQA uses F1; WideSearch uses Item F1 Max@4. † denotes reproduced results in the report’s evaluation setup.
| Model | BrowseComp | BrowseComp-ZH | DeepSearchQA | LiveBrowseComp | HLE | WideSearch |
|---|---|---|---|---|---|---|
| Nex-N2-Pro | 83.7 | 79.6† | 92.3† | 50.4† | 50.0† | 75.6 |
| MiroThinker 1.7 | 74.0 | 75.3 | — | 34.1† | 42.9 | — |
| apodex-1.0 | 75.5 | 82.6 | 84.6 | — | 49.0 | — |
| XYZ-Aquila-pro | 84.8 | 85.1 | 92.5 | 53.7 | 53.3 | 81.2 |
| apodex-h1 | 90.3 | 84.1 | 94.4 | — | 60.8 | — |
| Kimi-K2.6 (1T) | 83.2 | — | 92.5 | 31.7 | 55.5 | 80.8 |
| GPT-5.5 xhigh | 84.4 | — | — | — | 52.2 | — |
AI4AI uses agents to improve an agentic system, but does not hand over the objective. Humans define goals, proxy benchmarks, resource envelopes, risk boundaries, and acceptance criteria. AI agents propose and execute scoped interventions, while the evaluator returns metrics and diagnostics without exposing answer keys or private labels.
The loop is recursive, but not open-ended. Accepted and rejected attempts both become auditable experience, turning development memory into an asset rather than a pile of disconnected experiments.
A bounded AI4AI loop keeps optimization broad enough to improve the system and narrow enough to inspect.
Each released agent combines a selected backbone, graph-grounded data construction, state-faithful supervised post-training, a fixed three-tool execution protocol, context management, replay infrastructure, and a versioned evaluation gate.
AI4AI steers task construction toward questions that can be solved through observable search behavior, not memorized answers. The answer-hidden contract keeps rollout agents away from evaluator-side labels.
Training examples are aligned to the exact visible state at each turn, preserving the decisions the agent actually faced while supporting long-sequence packing and context-managed rollouts.
The agent can call search, scrape, and python under fixed semantics, visible-context limits, and turn budgets. Runtime and replay improvements are allowed only when they improve observability or execution without strengthening the task interface.
Retrieval-heavy tasks emphasize search; GAIA and DeepSearchQA use more extraction and computation. Strong search agents need flexible tool allocation, not a fixed interaction template.
The released XYZ-Aquila-mini and XYZ-Aquila-pro systems include the adopted data, SFT, runtime, context, and replay interventions. A process-reward RL design is screened as a promising extension, but it is not counted as part of the reported release because it has not completed the full training-and-gate path.
External benchmarks stay outside the optimization state and are used for final or explicitly authorized evaluation.
Candidate gains cannot come from extra tools, larger visible context, or larger turn budgets.
Every supervised or analyzed decision is tied to the exact visible state available at that step.
The technical report is explicit about remaining work: finer ablations are needed to quantify each intervention class; cost and latency should be reported more completely; live web benchmarks are dated observations; and the RL layer needs broader validation. The same discipline that produced the current systems should guide the next phase: define the scope, run the gate, preserve the evidence, and report what changed.