Three ADAS configurations:
an evidence review

Executive findings

The most useful result of this review is a narrower comparison unit. A model name alone does not settle the trim, option package, production period or installed software. We therefore keep selected-trim values separate from related platform and configuration-table evidence.

1. M9 has a bounded configuration mapping. Four fields have selected-configuration support, including one derived count. Their scope is explained in the case study below. [M9-SCOPE-02] [M9-001] [M9-002] [M9-003]
2. L6 remains a research target. Refreshed AD Max and L Series statements supply useful leads, but this review has not tied them to a dedicated 2025 refreshed L6 Max configuration record. Its selected values remain unknown. [L6-001] [L6-002] [L6-003]
3. G7 needs equipment and production-period mapping. A named 602 Max Flagship column is established; assigning hardware to that column still requires the applicable equipment and option markers. The base-table rating and Ultra-package ratings remain contextual evidence. [G7-SCOPE-02] [G7-001] [G7-002]

Across all three selections, the inspected material does not establish installed software or market/build-specific urban and highway navigation-assist rollout. “Unknown” records that evidential limit; it does not mean the feature is unavailable.

1. The comparison unit

All three records concern the China market. The manufacturing period is unresolved for each. A web configuration page is a revisable document: its inspection date is not a vehicle's build date.

The G7 target excludes Ultra and Ultra SE packages. L6's option package is not established. A precise research target name does not itself confirm that target's equipment.

Method and evidence standard

We inspected five unique manufacturer or supplier sources and recorded 13 claim-level evidence entries. Each entry preserves its URL, source location, publication date where available, inspection date and limitations. The field table and exports share one frozen JSON dataset.

We accept an exact-trim value only when the cited document maps it to the selected configuration. Family-level statements and unresolved table entries stay separate. A count obtained by addition is labeled as derived. Missing evidence remains null in JSON and blank in the CSV value column; neither encoding means zero.

No road test, teardown or vehicle-software inspection was performed. This pass is not an exhaustive inventory of every possible public disclosure.

2. What the three cases teach

AITO M9: a document mapping with a limited reach

The selected five-seat EREV Max+ sits within eight reviewed 2026 columns. Each repeats Huawei ADS 5 Advanced, the same lidar allocation and five millimetre-wave radar units. That explicit repetition supports this selection without relying on missing symbol-only equipment markers. [M9-SCOPE-02]

The lidar allocation is one 896-line dual-optical-path unit, one cabin-mounted common-optical-path unit and four solid-state units: six by addition. Cabin placement alone does not identify a monitoring purpose or all-unit supplier. The maker's radar count includes a distributed system; we do not infer its internal physical sensor count. [M9-002] [M9-003]

The inspected AD section did not establish compute SKU, chip count or TOPS. The system tier does not verify an installed build or a specific feature release. [M9-004] [M9-001]

Li Auto L6: use platform statements as leads

Li Auto's Q1 2025 results describe a Thor-U upgrade for refreshed AD Max. Hesai's May 2025 announcement describes ATL adoption in the refreshed L Series. These are platform and series statements in this evidence set, not a dedicated L6 Max equipment row. [L6-001] [L6-002]

Consequently, the selected-trim chip, count, rating and lidar fields remain unknown. A count cannot be inferred from singular wording, nor can a specification for another model or the earlier L6 establish the refreshed Max's value. [L6-003]

XPENG G7: preserve option and workload boundaries

The manufacturer names the 602 Max Flagship variant. The base hardware entry lists one Turing AD chip at 750 manufacturer-claimed effective TOPS, but this review has not verified the full per-trim equipment markers and applicable production period. Those values therefore remain related evidence. [G7-004] [G7-001] [G7-SCOPE-02]

The Ultra upgrade separately lists two AD chips at 1500 effective TOPS and a cockpit chip at 750 effective TOPS. Adding the workloads would mislabel cockpit capacity as AD compute. The package also must not be silently assigned to the base research target. [G7-002]

The inspected extracted hardware list does not provide an explicit applicable zero-lidar assertion. Omission of a row is insufficient to fill the field with zero. [G7-003]

3. All selected-trim fields

Related evidence above does not populate this table. Open the annotated comparison for each field's status, limitations and evidence IDs.

Hardware, announcements and rollout answer different questions

A hardware entry supports what the document lists. A feature announcement supports what was announced. A rollout claim needs evidence tying a release to a market, build, eligible configuration and deployment scope. An installed vehicle adds another verification step. None of those stages should be substituted for another.

All nine build/urban-rollout/highway-rollout fields are unknown in this dataset. A future update should say whether it adds an announcement, a limited release, a wider rollout or an observed installation, with its own source and date.

Why this report does not produce a hardware leaderboard

The unresolved trim mappings alone prevent a complete like-for-like hardware comparison. In addition, a manufacturer-claimed effective TOPS value is not a measured driving outcome. This review has no common workload, precision, software build or road-test protocol establishing a cross-vendor ranking. Chip and sensor quantities are retained as bounded specifications.

4. What would change the conclusion?

The next useful evidence should close a specific gap rather than add another unspecific headline.

• L6: a dedicated 2025 smart-refresh Max configuration record, with clear chip/sensor fields and applicability.
• G7: explicit 602 Max Flagship equipment and option markers, plus the production-period rule that applies to them.
• All three: the installed build and feature-release evidence specifying market, eligible configuration and rollout scope.

For a procurement or engineering comparison, retain the exact variant and build-period record, confirm installed equipment, capture the software build and feature status, then define a common test protocol. These are suggested verification steps, not tests performed for this report.

Using and citing the exports

Filter the CSV by record ID and field. Read the value together with its status, scope and evidence IDs; a related value belongs to its own related scope. Follow the original document before applying a value outside this snapshot. Keep dataset version 0.2 and the hash below with any derived comparison.

5. Sources, files and actual changes

1. HIMA: 2026 M9 configuration. Publication date not stated. Checked 18 October 2026.
2. Li Auto: Q1 2025 financial results. Published 29 May 2025. Checked 18 October 2026. Official English version.
3. Hesai: refreshed L Series ATL announcement. Published 9 May 2025. Checked 18 October 2026.
4. XPENG: 2026 G7 configuration. Publication date not stated. Checked 18 October 2026.
5. XPENG: 2026 product launch. Published 8 January 2026. Checked 18 October 2026.

The complete 13-entry evidence register contains source locations and field-level limitations omitted from this narrative. Source accessibility can vary; the linked Li Auto English equivalent was separately checked on 21 October.

Changes from v0.1 to data v0.2: M9 was narrowed to the named five-seat EREV Max+ for four reviewed fields. G7 was narrowed to a named target while its hardware stayed unknown. L6 family evidence remained separate. The normalized dataset now supplies matching HTML and CSV. These are scope and delivery changes, not observed changes to vehicles.

This report: adds interpretation and a verification plan to that same v0.2 dataset. It adds no confirmed hardware values and does not replace the dated two-page sample.

Dataset SHA-256:
b39139094a68f59ffaf5ecf69d097d98d7cd32614a64018ce39ebc938694742b