[REVISION HISTORY]

OpenAI GPT-6 model development and legacy code migration

Updated 3 times since CLSTR started tracking revisions of this situation.

What changed

2026-10-21 13:35 UTC → 2026-10-22 09:44 UTC · added ~~removed~~

OpenAI is preparing to release new AI models by the end of the year, potentially utilizing Looped Transformer technology to improve performance in mathematics and coding while reducing costs. These developments are subject to safety evaluations and government reporting requirements under executive orders. Subsequent updates indicate significant advancements in the models' ability to handle legacy programming languages. Specifically, the ~~GPT-6 1~~ GPT-6.1 Sol model has demonstrated up to 97.9 percent accuracy in migrating legacy programs to modern languages such as C++, Java, Python, or Rust. This represents a massive leap from ~~GPT-5 6~~ GPT-5.6 Sol, which recorded only 5.4 percent accuracy in similar low-effort reasoning tasks. In migration tests designed to translate COBOL into modern languages while preserving exact functional behavior, ~~GPT-6 1~~ GPT-6.1 Sol achieved 86.6 percent accuracy at low reasoning effort. Other models, including GPT-6 Astra, have shown notable progress in handling low-resource languages like COBOL, Fortran, and Visual Basic. In the MultiLangBench evaluation, which requires building SQLite-compatible database engines from scratch across 20 languages, GPT-6 Astra achieved an average success rate of 87.4 percent, up from 26.8 percent for ~~GPT-5 6~~ GPT-5.6 Sol. Specifically, GPT-6 Astra reached success rates of 72.1 percent in COBOL, 83.5 percent in Fortran, and 89.4 percent in Visual Basic. These improvements target languages critical for global infrastructure, such as COBOL, which supports an estimated 40 percent of online banking and 95 percent of ATM transactions. Additionally, the GPT-6 Luna model has been introduced as a cost-effective option, offering an estimated 78 percent reduction in API costs compared to previous models.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.