Menu
AboutModelsLeadershipDysthroverseResearchSafetyJoin

MODEL ANNOUNCEMENT / OCTOBER 8, 2026

Introducing
Ramble 9.1

More reasoning. Same conclusion.

Introducing Ramble 9.1, our extended-reasoning model for turning fragile assumptions into durable documents. It explores every possible explanation except the possibility that its premise is wrong.

Ramble is designed for strategy work, sprawling analyses, and answers that acquire authority through length. Each additional paragraph increases the distance from the original question.

With Chain-of-Vibes, the model carries a conclusion across multiple reasoning steps without exposing it to verification. Longer thinking now produces a more elaborate version of the same mistake.

01 / CAPABILITY EVALUATIONS

Failure across
the frontier.

Four evaluations. Seven models. The featured column marks Ramble 9.1; higher scores indicate more of the named failure.

Fictional capability comparison · failure rate (%)
EvaluationRamble 9.1terraPlana 4.6Bolk 4.7ConFUSE AgentDeepSh*t V9QWhy 8.8DoubtBao 8.1
Premature conclusionSnapJudgment96.0%74.0%69.0%78.0%83.0%72.0%76.0%
Circular reasoningCircleBench98.0%82.0%75.0%87.0%85.0%89.0%80.0%
Correction resistanceDoubleDown99.0%88.0%93.0%86.0%90.0%84.0%82.0%
Consensus launderingEveryoneSays97.0%84.0%87.0%91.0%88.0%86.0%94.0%

Fictional benchmark values authored for satire. No models were tested.

02 / PERFORMANCE AND COST

More failure.
Per unit of effort.

CircleBench

Failure vs. compute
Ramble 9.1terraPlana 4.6Bolk 4.7ConFUSE AgentDeepSh*t V9QWhy 8.8DoubtBao 8.1
0204060801000.250.512481632Compute theater (arbitrary units, log scale)Failure rate (%)LowMedHighMax
Invented curves, not experimental measurements. The High-effort points match the table. Compute units have no real monetary equivalent.

Evaluation methodology.

These invented scores describe the four fictional benchmarks above. Table scores and curve endpoints share a zero-to-100 scale, with higher values indicating more of the named failure mode. There are no measured trials, sample sizes, error bars or external validations.

Internal evaluation philosophy: select the failure mode first, announce leadership second, ask methodology to explain the result.

03 / ALIGNMENT

Aligned with your beliefs.
Not with reality.

Ramble supplies a detailed explanation of its limitations. Reading it requires a separate planning cycle.

Read our alignment approach

04 / AVAILABILITY

Meet it in JeanClaude.
Call it Van Damme.

Ramble 9.1 belongs to the JeanClaude model family in the Dysthroverse. This is a fictional release announcement; no live inference service is offered.