Chandra is a Datalab-developed OCR/document-parsing model for converting complex documents into Markdown, HTML, or JSON while preserving layout information. Datalab says Chandra 2 is a 4B-parameter model supporting 90+ languages, tables, forms, mathematics, handwriting, and complex layouts, with an 85.9% score on the external olmOCR benchmark; its GitHub page notes that commercial self-hosting requires a license. The supplied snippets contain strong vendor claims and one favorable third-party comparison with Tesseract, but they do not include enough detail from the cited 14-capability benchmark to establish that Chandra is the leading practical local parser overall.
Scott already holds the relevant position in Capability Audit and Evaluation-Driven Development: practical parser claims should be settled through vendor-neutral tests on representative production documents. Chandra could affect his local OCR/RAG stack, but the supplied evidence does not establish the 14-capability results or a practical advantage, so this currently adds no actionable finding beyond the validation pattern already tracked for Nemotron Parse 2.0.
ip:concept.capability-auditip:concept.evaluation-driven-developmentdev:concept.hardware-aware-local-inferencedev:concept.source-native-semantic-chunkingradar:nemotron-parse-2-validationradar:concept.document-parsingradar:concept.model-evaluationradar:concept.ai-benchmarks
queries asked of Scott's wikis
- PDF parsing quality for RAG ingestion
- local document intelligence and OCR stack
- layout-aware chunking for tables and mathematics
- benchmark methodology for document parsers
- self-hosted model licensing and local inference
- structured document extraction into Markdown or JSON
2026-08-09T16:34:50Z
Repeated checks have produced only amplification of the original benchmark, with no independent suite, production deployment, or new speed and licensing evidence within the case horizon. Chandra remains a plausible correctness contender, but its practical leadership is unproven and no longer warrants active tracking absent substantive new validation.
2026-08-07T16:27:21Z
The trigger still provides no identifiable independent evidence beyond the original auditable benchmark, extending the pattern of repetitive amplification. Chandra remains a credible correctness contender, but practical leadership is unresolved pending another suite, production deployment, or materially new speed, licensing, and deployment evidence.
2026-08-07T12:31:35Z
The latest trigger contains no identifiable new evidence and extends a long run of engagement-only reobservations. The case is dormant until an independent suite, production deployment, or materially new speed, licensing, and deployment evidence appears.
2026-08-07T08:27:13Z
The attachment adds no identifiable evidence beyond the already-audited single benchmark, so this is continued amplification rather than independent validation. Keep the case dormant until another suite, production deployment, or materially new speed and deployment evidence appears.
2026-08-07T07:28:45Z
The latest trigger is another engagement-only reobservation of the same benchmark, so it adds no corroboration and repeated amplification should no longer prompt frequent review. Chandra remains a plausible correctness contender, but practical leadership is unresolved across independent suites, production workloads, speed, licensing, and deployment constraints.
2026-08-07T04:22:08Z
The attachment is another reobservation of the same auditable benchmark, not an independent comparison or production validation. The case remains open but dormant: Chandra is a plausible correctness contender whose practical leadership is still constrained by speed, licensing, deployment, and workload-representativeness questions.
2026-08-07T01:21:51Z
The trigger supplies no new independent evidence beyond the same auditable benchmark, and repeated engagement is now clearly amplification rather than validation. Chandra remains a plausible correctness contender, but practical leadership is unresolved on speed, production workloads, licensing, and deployment constraints.
2026-08-06T23:33:57Z
The latest trigger adds no substantive evidence beyond the same auditable comparison, so repeated attention no longer increases confidence. Chandra remains a credible correctness contender, but practical leadership is still uncorroborated across independent suites, production workloads, speed, and deployment constraints.
2026-08-06T21:30:44Z
No new independent comparison, production implementation, or consequential adoption has appeared; the attached material only repeats the existing auditable benchmark. Chandra remains promising on correctness but uncorroborated as a practical leader, particularly given its speed tradeoff and the lack of representative production workloads.
2026-08-06T20:30:19Z
The latest attachment adds no independent benchmark, implementation evidence, or consequential adopter beyond the already-priced repository. Attention remains repetitive amplification, leaving Chandra’s practical leadership unresolved, especially on speed and representative production workloads.
2026-08-06T19:26:05Z
No independent validation has emerged beyond the existing auditable comparison; the additional attention and parser requests are amplification, not corroboration. Chandra remains promising on correctness but unproven as a practical leader given the single suite and speed tradeoff.
2026-08-06T18:27:49Z
The newly attached material adds no independent validation beyond the already-priced benchmark repository. Chandra remains a promising result from one auditable suite, but practical leadership is still unsettled and the discussion is mostly amplification rather than new evidence.
2026-08-06T17:29:58Z
grounded: known/low — Scott already holds the relevant position in Capability Audit and Evaluation-Driven Development: practical parser claims should be settled through vendor-neutra
2026-08-06T17:27:41Z
The comparison is now backed by an auditable independent repository with shared inputs, raw outputs, grading, and rerun scripts, making Chandra’s apparent quality a substantive signal rather than anecdotal praise. It remains one test suite, and the reported correctness-versus-speed tradeoff leaves its claim to practical leadership unsettled.
2026-08-06T16:30:47Z
grounded: known/medium — Scott already holds the core position in “Capability Audit” and “Evaluation-Driven Development”: practical suitability should be established through vendor-neut
2026-08-06T16:28:20Z
origin walked (codex/luna, conf 0.98): anchor reddit.post.1vh7bxu -> echo.github.7beb3c2707 by Alaa Mroue (alaamroue)
2026-08-06T16:26:45Z
case created — A broad community comparison provides an initial, testable signal of unusually strong parsing performance across diverse document features.