Chandra 2.7

Chandra 2.7 is the document-parsing model version Datalab said went live on 12 November 2026. In the official blog Chandra 2.7: better tables and languages, dated November 12, 2026, Hunter Heidenreich wrote: “Chandra 2.7 is live in the Datalab API, and balanced and accurate modes run on it (fast mode runs Surya 2.6). You don’t need to change anything to use it, and the speed and price are the same.” This site’s sitemap scan shows that post as a new datalab.to URL since the previous 00:32 pass.

According to the company, the main gains are in tables, non-English pages, and rotated pages. On its public parsing benchmark OmniParseBench (16,288 pass/fail tests on 2,937 pages), official table scores rose from 91.6 to 94.2 in balanced mode and from 94.2 to 95.3 in accurate mode. Official overall scores are 94.4 for accurate and 93.5 for balanced. On the 1,935 tests across 534 pages tagged as non-English, the official pass rate rose from 86.7% to 87.9% in accurate mode and from 82.7% to 83.0% in balanced mode. On rotated pages, Datalab says balanced-mode pass rates rose from 81.4% to 87.3%. The post also says math pages and degraded scans score about 1 point lower, with multi-column pages about the same. Those figures are company results from a 12 November 2026 production-API run; some Gemini 4 comparison rows include a rerun note. This page does not independently reproduce them.

Pricing is officially unchanged from Chandra 2.6: balanced is $4 per 1,000 pages ($3 with data sharing), and accurate is $10 ($7.50 with data sharing). Chandra 2.7 is live in the API and the playground; for on-premises deployments, the post says to contact [email protected].

Key points:

• Official blog: https://www.datalab.to/blog/chandra-2-7 (12 November 2026);
• This is a concrete Datalab model-version name. Do not confuse it with the existing entry for Chandra 2.5, with the earlier Chandra 2.6 release, or with Surya 2.6, which official materials say still powers fast mode.

Why it matters

This is a version name Datalab announced on 12 November 2026 in an official blog that says the model is already live in the production API — not a third-party rumor. The post first appeared in this site’s sitemap during this scan window. For anyone tracking document-parsing model iterations, Chandra 2.7 is distinct from the already recorded Chandra 2.5 entry.

This page only restates official public material. The OmniParseBench scores, non-English and rotated-page pass rates, and comparison rows against Gemini 4, GPT, Claude and others are company statements. This page does not independently reproduce them.

Limits

This is an incremental version in an existing product line, not a new brand. Official materials also say math pages and degraded scans scored slightly lower, and that fast mode still runs Surya 2.6. The public post does not include a full, independently downloadable item-by-item reproduction pack. Do not treat this entry as a guarantee of production OCR or table-parsing quality.

1. Read the official blog at https://www.datalab.to/blog/chandra-2-7 , and cross-check the current Datalab API / playground docs and pricing page;
2. Follow the official note that balanced and accurate modes run Chandra 2.7 with no call-site change, while fast mode still uses Surya 2.6; for on-premises deployments, contact [email protected] ;
3. When citing OmniParseBench, non-English or rotated-page scores, or per-1,000-page prices, attribute them to the 12 November 2026 official blog and recheck the upstream source;
4. Do not describe Chandra 2.7 as Chandra 2.5, Chandra 2.6, or Surya 2.6; this page only restates the official blog — recheck the upstream source before citing it externally.