Astra's reported reasoning gains with filler tokens
Redwood Research reports that GPT-6 Astra performs better with filler tokens: modestly on general benchmarks and significantly on serial depth-heavy tasks.
TLDR
Redwood Research reports that GPT-6 Astra performs modestly better on general benchmarks with filler tokens and significantly better on serial depth-heavy tasks, unlike other models it tested. A user also says Astra can use both user-provided filler tokens and tokens it generates itself.
Combined views
—
6 Sources, first seen ago
