Technical blog · July 2026

AI4AI at Scale: Building Open-Weight Deep Search Agents

XYZ-Aquila is a family of search agents developed through a bounded, verification-gated AI-for-AI loop: humans set the objective and acceptance gate, while AI agents diagnose failures and implement scoped changes to data, post-training, runtime, context management, and infrastructure inside that envelope.

26-page slim technical report July 22, 2026 2.8 MB PDF
Bar charts comparing XYZ-Aquila-mini and XYZ-Aquila-pro with search agents on six benchmarks
Overview across six of the seven evaluated agentic search benchmarks.
01 · Results

Strong matched-scale results for Deep Search.

XYZ-Aquila-mini and XYZ-Aquila-pro are evaluated on BrowseComp, BrowseComp-ZH, DeepSearchQA, GAIA, LiveBrowseComp, Humanity’s Last Exam, and WideSearch. The external suite is withheld from routine optimization; development decisions use a private, answer-hidden gate. These results evaluate the final systems, not the causal contribution of any single intervention.

78.8%
BrowseComp accuracy for XYZ-Aquila-mini.
<40B table leader
89.5%
DeepSearchQA F1 for XYZ-Aquila-mini.
evidence aggregation
85.1%
BrowseComp-ZH accuracy for XYZ-Aquila-pro.
<400B open-weight leader
81.2%
WideSearch Item F1 Max@4 for XYZ-Aquila-pro.
broad retrieval
Small-scale open-weight agents

In the reported <40B open-weight table, XYZ-Aquila-mini leads every benchmark column.

ModelBrowseCompBrowseComp-ZHDeepSearchQAGAIALiveBrowseCompHLEWideSearch
Agents-A175.596.029.6†47.6
Nex-N2-mini74.179.6†87.2†41.4†37.1†62.0
apodex-mini71.580.682.232.8†46.8
MiroThinker 1.7 mini67.980.334.9†36.473.3†
XYZ-Aquila-mini78.882.989.597.148.751.180.8

This is a column-wise comparison: no baseline reports all seven benchmarks. DeepSearchQA uses F1; WideSearch uses Item F1 Max@4. † denotes results reproduced in the report’s evaluation setup; unmarked baselines come from public reports or benchmark submissions.

Large-scale comparison

XYZ-Aquila-pro leads the listed <400B open-weight group and remains competitive in broader context.

ModelBrowseCompBrowseComp-ZHDeepSearchQALiveBrowseCompHLEWideSearch
Open-weight agents (<400B parameters)
Nex-N2-Pro83.779.6†92.3†50.4†50.0†75.6
MiroThinker 1.774.075.334.1†42.9
apodex-1.075.582.684.649.0
XYZ-Aquila-pro84.885.192.553.753.381.2
Frontier-scale and proprietary agents
apodex-h190.384.194.460.8
DeepSeek-V4-Pro (1.6T, Max)83.438.3
Kimi-K2.6 (1T)83.292.531.755.580.8
Claude Opus 4.779.389.154.7
GPT-5.5 xhigh84.452.2
Qwen3.6-Plus74.3

The <400B open-weight group is the primary matched-scale comparison; the second group is broader context. Public, unmarked values come from different evaluation stacks and dates and are not uniformly controlled. DeepSearchQA uses F1; WideSearch uses Item F1 Max@4; † denotes reproduced results.

02 · AI4AI

Recursive improvement, bounded by design.

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 approved diagnostics without exposing answer keys or private labels. Each candidate receives at most one gate query at the cycle boundary.

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.

Bounded AI4AI loop connecting human contracts, agent-led diagnosis and interventions, an isolated evaluation gate, and shared experience memory

A bounded AI4AI loop keeps optimization broad enough to improve the system and narrow enough to inspect.

01
Set the contractDefine target capability, benchmarks, tools, budgets, and acceptance gates.
02
Diagnose failuresAgents analyze traces, logs, reviewer comments, costs, and reliability signals.
03
Propose changesCandidates touch data, SFT, runtime, context policy, infrastructure, or screened RL.
04
Gate adoptionAn isolated evaluator measures candidates without revealing private labels or answer keys.
05
Bank experienceAccepted and rejected attempts are retained with provenance and decision status.
03 · System

XYZ-Aquila is a system configuration, not only a model checkpoint.

XYZ-Aquila-mini uses the Qwen3.6-35B-A3B backbone, while XYZ-Aquila-pro uses Qwen3.5-397B-A17B. Each released agent combines its backbone with graph-grounded data construction, state-faithful supervised post-training, a fixed three-tool execution protocol, context management, replay infrastructure, and a versioned evaluation gate.

Pipeline for constructing graph-grounded, tool-reachable search tasks while keeping answers evaluator-side
Data

Graph-grounded and tool-reachable tasks.

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.

State-faithful supervised fine-tuning pipeline built from exact agent-visible states
Post-training

State-faithful SFT for long traces.

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.

Execution harness preserving fixed search, scrape, and Python tools with exact replay state
Runtime

Fixed interface, exact replay.

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.

Stacked bars showing different search, scrape, and Python call shares across six benchmarks
Behavior

Adaptive tool use emerges across benchmarks.

Search makes up 78–84% of sampled calls on BrowseComp, BrowseComp-ZH, and LiveBrowseComp; extraction and Python together account for 54–55% on DeepSearchQA and GAIA. Strong search agents need flexible tool allocation, not a fixed interaction template.

04 · Boundaries

What is adopted, what is screened, and what remains open.

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.

Three principles keep the loop honest.

Isolation.

External benchmarks stay outside the optimization state and are used for final or explicitly authorized evaluation.

Invariance.

Candidate gains cannot come from extra tools, larger visible context, or larger turn budgets.

Auditability.

Every supervised or analyzed decision is tied to the exact visible state available at that step.

Limitations and future work

The report supports a bounded system-optimization claim under a specific contract. It does not yet establish which components are causally necessary, whether development is resource-efficient end to end, or whether the recipe transfers beyond search.

Causal attribution

The study evaluates endpoint systems, not a complete decomposition of path-dependent data, training, runtime, and context interventions. Controlled retraining and targeted factorial ablations are still needed.

Adaptive overfitting

Evaluator isolation reduces but cannot eliminate adaptation to the private development distribution. Rotating splits, refreshed hidden holdouts, and scheduled transfer audits remain important.

Comparability and cost

Some baselines come from different tools, judges, dates, and live-web conditions. The report also lacks end-to-end compute, latency, tool, evaluator, and human-review accounting.

RL and transfer

The process-reward RL design passed low-cost screening but is not part of either release. Empirical validation remains limited to Deep Search and two backbones.