Research noteResearched 20 Oct 2026 · published 20 Oct 2026, 12:51 IST

Research note: OpenAI says its newest models handle COBOL and Fortran far better, and releases a 20-language coding test

How we checked OpenAI’s coding results and why we label the scores as company claims.

Research led by Mayank Sahu

Compiled with AI assistance (Claude by Anthropic)

Read the story →

At a glance

2searches run

3sources read

2sources used

6facts labelled

Mixedevidence strength · 3 of 6 facts (50%)

How we rate evidence: Some facts are confirmed, others rest on one report or on the company’s own claims. We count the share of known facts labelled Confirmed or Reported: 70% or more is Strong, 40–69% Mixed, below 40% Limited.

From sources to story

How many sources we read, how many we used, and how many are official or archived.

Read3Every source we opened

Used2Relied on for the story

Cited2Linked in the story

Official1From the company or a public body

Archived2Saved on the Internet Archive

How sure we are, fact by fact

321

Confirmed · 3 — An official source says it: lab docs, pricing page, notice.
Company claim · 2 — Only the company says it. Not checked by anyone else yet.
Unconfirmed · 1 — One report only, shown with its source.

Evidence map

Each row is a fact from the story; each column is a source. A filled dot means that source supports the fact.

Fact	Label	1OFF	2OUT
OpenAI published results on low-resource programming languages on 19 October 2026.	Confirmed
MultiLangBench asks models to build a SQLite-compatible engine in 20 languages.	Confirmed
GPT-6 Astra averaged 87.4% on MultiLangBench, up from 26.8% for GPT-5.6 Sol.	Company claim
GPT-6.1 Sol scored 86.6% (low effort) and 97.9% (higher effort) on COBOL migration.	Company claim
MultiLangBench is released on GitHub.	Confirmed
The COBOL migration test is not public.	Unconfirmed

Question

What we set out to check

What OpenAI measured, the exact scores, whether the tests are public, and what limits OpenAI states.

Findings

What we found

OpenAI’s post gives all scores and describes both tests.
The CellCog summary matches every figure and notes the migration test is not released.
No independent group has published results on these tests yet.

Discrepancies

Numbers that differ

None between sources. CellCog describes GPT-6.1 Sol’s cost as about 61% of GPT-5.6 Sol’s; OpenAI says about 39% less. Both describe the same figure.

Method

Labels we chose and why

Claim for all scores, because OpenAI built and ran the tests itself.
Confirmed for what the tests are and the GitHub release.

Open questions

What we could not verify

Independent reproduction of the scores.
The migration test itself.

Can you help confirm one of these? Send us a source. When we find an answer, we update this note and the story, with the date.

Next steps

How we could improve this

Our second source is a company blog, which is weaker than an independent newsroom. We will add an independent report or third-party results when available.

Research timeline

Research20 Oct 2026
Research note published20 Oct 2026, 12:51 IST
Story published20 Oct 2026, 13:01 IST

What we searched

OpenAI COBOL Fortran MultiLangBench GPT-6
OpenAI low-resource programming languages COBOL migration GPT-6.1 Sol 97.9%

Every source we read

Source	Used?	Why
OpenAI Archived copy	USED	Primary source for tests and scores.
CellCog blog Archived copy	USED	Summary with caveats; the author sells AI coding tools, so used only for points that match OpenAI’s post.
FourWeekMBA	NOT USED	Could not open.

Sources cited in the story

Research notes show our working. They are not peer-reviewed papers. Data: facts as JSON · How we report