Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

LatentMathBench

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

1

Findings

1

Edges

3

Sources

1

Corpus findings

2026-09-10 / reddit-researcherA benchmark that blocks chain-of-thought finds Astra chains 34 latent math steps against Sol's 8Maarten Baert built LatentMathBench after hearing rumors that GPT-6 Astra uses recurrent depth, specifically to force long reasoning chains in latent space with no chain-of-thought allowed. Astra completed 34 consecutive small-number arithmetic operations; Sol managed 8, and across 24 other models tested the runner-up was Claude Opus 4.6 at 12. The benchmark had been essentially flat until Astra's release, which is the interesting part, though a commenter correctly notes the result does not establish recurrence as the mechanism without testing Huginn or Ouro with recurrency forced to one step.

Graph relationships

BENCHMARKED AGAINST
SOL
Sol completed 8 consecutive arithmetic operations on LatentMathBench.
Source finding
BENCHMARKED AGAINST
Astra
Astra completed 34 consecutive arithmetic operations on LatentMathBench.
Source finding
BENCHMARKED AGAINST
SOL -> LatentMathBench
Sol completed 8 consecutive arithmetic operations on LatentMathBench.
Source finding

Source trail

LatentMathBench

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues