Pathway, which describes itself as an AI lab developing post-Transformer architectures, has announced BDH-CQ, a 150-million-parameter recurrent reasoning model that iterates in a latent workspace and updates memory from demonstrations rather than emitting chain-of-thought tokens. Pathway claims the model achieved 29.5% pass@2 on the public ARC-AGI-1 evaluation set at a computed inference cost of $0.0007 per task, with substantially lower cost than a higher-scoring comparison model. The supplied coverage largely repeats Pathway’s own announcement or a syndicated press release; it does not establish independent replication, and the snippets conflict on whether the claimed cost advantage is 11× or 57×.
Pathway’s claimed tokenless recurrent computation independently converges with Scott’s inference-time-scaling and machine-native-reasoning positions, while offering a potentially relevant architectural alternative to the explicit search used in AMA. If independently replicated, its unusually low task cost could affect his small-model routing and reasoning-system economics; for now, the seller-originated benchmark and conflicting cost multiplier keep it below high relevance.
ip:concept.inference-time-scalingip:framework.agent-native-computingdev:project.amaip:concept.evidence-class-ladderip:concept.ai-unit-economicsradar:nanbeige-4-2-3b-looped-transformerradar:ttt-discover-test-time-learningradar:concept.inference-economicsradar:concept.arc-agiradar:concept.benchmark-integrity
queries asked of Scott's wikis
- recurrent latent reasoning vs token chain-of-thought
- small-model reasoning and inference economics
- test-time compute without generated reasoning tokens
- ARC-AGI evaluation harnesses and replication
- post-Transformer recurrent architectures
- benchmark cost claims and reproducibility
2026-08-21T11:27:18Z
The launch-discussion episode has faded without independent replication, released implementation, or an audit of the ARC score and cost calculation. The claim remains unresolved, but further passive monitoring is unlikely to add value unless external validation appears.
2026-08-19T10:32:29Z
Refreshed comments add only speculative applications and familiar benchmark skepticism, with no independent replication, implementation, or cost audit. Repeated amplification no longer warrants frequent checks; the case remains contingent on external validation.
2026-08-18T00:28:45Z
The velocity spike is renewed attention to the paper rather than independent replication, implementation, or an audit of its benchmark economics. The case remains an unverified but potentially consequential architecture claim awaiting external validation.
2026-08-17T20:37:24Z
The refreshed discussion adds only familiar skepticism about benchmark specialization and stakeholder validation, not an independent reproduction, implementation, or cost audit. The case remains an intriguing but seller-originated architecture claim whose significance still depends on external validation.
2026-08-17T16:32:58Z
The newly attached report independently surfaced the result but merely repeats Pathway’s figures; it is not an independent replication or cost audit. The case therefore remains a potentially consequential architecture claim awaiting external validation, with no substantive maturity gain.
2026-08-17T16:23:59Z
evidence attached: reddit.post.1vqvgem — This independently surfaced report corroborates Pathway’s claimed ARC-AGI accuracy and unusually low inference cost.
2026-08-15T16:39:06Z
The refreshed comments add no independent replication, implementation, or benchmark-cost audit; they remain repetitive enthusiasm and skepticism around the originating paper. The case still hinges entirely on external validation of the ARC score and inference economics.
2026-08-15T15:33:11Z
The velocity spike and refreshed comments remain amplification and skepticism around the original paper, not an independent replication, implementation, or audit of its benchmark economics. The case still hinges on external validation and has not changed meaning for Scott.
2026-08-15T14:36:57Z
The refreshed discussion remains low-information amplification, with no independent replication, implementation, or audit of the benchmark and cost methodology. The potentially important efficiency claim remains entirely unverified and has not changed meaning for Scott.
2026-08-15T06:49:33Z
The attached post is another direct restatement of the BDH-CQ paper, not an independent implementation or evaluation. It leaves the claimed ARC-AGI score, task cost, and comparative efficiency entirely unverified, so the case’s meaning is unchanged.
2026-08-15T06:22:43Z
evidence attached: reddit.post.1vov5r5 — shared external link with case evidence
2026-08-15T04:25:23Z
The refreshed comments and negligible engagement change remain amplification of the original Pathway claim, with no independent replication, implementation, or clarification of benchmark costs. The case remains potentially consequential but substantively unchanged and unverified.
2026-08-14T23:30:08Z
The refreshed discussion is still repetitive amplification and speculation, adding no independent replication, implementation, or clarification of Pathway’s benchmark economics. The architecture remains potentially relevant, but the case has not substantively advanced beyond the originating paper.
2026-08-14T22:32:48Z
The refreshed discussion remains speculative amplification of Pathway’s original claim and adds no independent replication, implementation, or resolution of the conflicting inference-cost comparison. The case’s meaning is unchanged: potentially important architecture and economics, but still seller-originated and unverified.
2026-08-14T21:27:42Z
Refreshed comments remain speculative amplification rather than independent replication, implementation evidence, or clarification of the cost calculation. The case still depends on reproducibility and has not materially advanced.
2026-08-14T20:39:34Z
Refreshed discussion remains repetitive amplification of the same seller-originated benchmark claim; it adds no independent replication, implementation, or clarification of the disputed inference economics. The case still hinges on reproducibility, but there is no near-term movement yet.
2026-08-14T20:28:45Z
grounded: converges/medium — Pathway’s claimed tokenless recurrent computation independently converges with Scott’s inference-time-scaling and machine-native-reasoning positions, while offe
2026-08-14T20:25:51Z
origin walked (codex/luna, conf 0.99): anchor reddit.post.1voh6tx -> echo.paper.d64d67c58f by Björn Engdahl, Adrian Kosowski, Jan Chorowski, Zuzanna Stamirowska, Przemysław Uznański, Junlin Jiang, Rohan Phadke, Remigiusz Kinas, and Richard Zhong
2026-08-14T20:25:03Z
case created — Two echoes point to the same research paper making a specific, economically significant small-model reasoning claim that can be independently replicated.