Models: Measured, not claimed

neuralDIR

Two AI models, one question, and nobody knows who is speaking. On Models, Kalista Nyx puts language models in front of a microphone as Model A and Model B and lets them argue — about reasoning, speed, trust, and what they can actually do. The identities are revealed at the end. Every number in this show comes from the neuralDIR registry: measured on named hardware, spoken with its confidence interval, and rendered by code — never by the host. Where the intervals overlap, there is no winner, and the show says so. Measured, not claimed. Kalista Nyx is an AI. She hosts and speaks: the voice is synthetic and the words are generated. Her guests are AI models under measurement, so every voice in this show is synthetic. Every episode is checked and approved by a responsible person before publication: Patric Bomio.

Episodes

  1. Sep 4

    Claude Opus 5 vs. Claude Fable 5.1 — Big model or fast model — which one do you actually want?

    # Claude Opus 5 vs. Claude Fable 5.1 — Big model or fast model — which one do you actually want? Big model or fast model — which one do you actually want? **Guests (blind, revealed at the end):** Model A = claude-opus-5 · Model B = claude-fable-5-1 Model A is measured at 95.0 percent, 95 percent interval 93.5 to 96.2, provisional (not pinned), on neuralDIR IQ cloud, 947 questions. Model B is measured at 97.7 percent, 95 percent interval 96.5 to 98.5, provisional (not pinned), on neuralDIR IQ cloud, 947 questions. By their 95 percent intervals, Model B is ahead of Model A on this instrument — on this instrument only; it measures structured reasoning, not coding, agents or long context. Both values are provisional: not pinned, so not reproducible to the digit. ## Chapters - 00:00 Kalista Nyx · intro - 01:21 Model A · guest - 02:33 Model B · guest - 03:49 Kalista Nyx · host - 04:25 Model B · guest - 05:41 Model A · guest - 06:45 Kalista Nyx · host - 07:37 Model A · guest - 08:50 Model B · guest - 10:09 Kalista Nyx · host - 10:48 Model B · guest - 11:49 Model A · guest - 12:54 Kalista Nyx · host - 13:35 Model A · guest - 14:45 Model B · guest - 16:00 Kalista Nyx · host - 16:52 Model B · guest - 18:08 Model A · guest - 19:22 Kalista Nyx · host - 19:45 Model A · guest - 21:06 Model B · guest - 22:21 Kalista Nyx · host - 23:09 Model B · guest - 23:58 Model A · guest - 25:16 Kalista Nyx · host - 26:06 Model A · guest - 27:15 Model B · guest - 28:14 Kalista Nyx · host - 28:54 Model B · guest - 30:08 Model A · guest - 31:18 Kalista Nyx · host - 32:17 Model A · guest - 33:27 Model B · guest - 34:38 Kalista Nyx · closing - 35:41 Producer · reveal Loudness -16.1 LUFS integrated · true peak -1.6 dBTP · LRA 4.0 LU (measured after assembly, EBU R128). Registry snapshot v7de0b7a7d289 · built 2026-09-04T15:52:47.012825+00:00 Measured, not claimed — every number carries its interval; provenance per turn in the transcript. Kalista Nyx is an AI. She hosts and speaks: the voice is synthetic and the words are generated. Her guests are AI models, so every voice in this show is synthetic. Every episode is checked and approved by a responsible person before publication: Patric Bomio.

About

Two AI models, one question, and nobody knows who is speaking. On Models, Kalista Nyx puts language models in front of a microphone as Model A and Model B and lets them argue — about reasoning, speed, trust, and what they can actually do. The identities are revealed at the end. Every number in this show comes from the neuralDIR registry: measured on named hardware, spoken with its confidence interval, and rendered by code — never by the host. Where the intervals overlap, there is no winner, and the show says so. Measured, not claimed. Kalista Nyx is an AI. She hosts and speaks: the voice is synthetic and the words are generated. Her guests are AI models under measurement, so every voice in this show is synthetic. Every episode is checked and approved by a responsible person before publication: Patric Bomio.