TL;DR
The short version
Anthropic's Opus 5.5 reclaimed early-adopter love on coding, visuals, and writing feel (with verified $4/$20 pricing and Artificial Analysis index 58), while OpenAI's Soul/Luna push cost-efficiency for high-volume iteration.
The useful frame is multi-model stack slots plus harness lock-in — personality is UX, and pacing can look like iterative mid-tier gains with real cost cuts.
Based on The AI Daily Brief (published 25 Sep 2026).
Two pitches, one Tuesday
Opus 5.5 is sold as a return after a weak Opus 5: lower list price than Opus 5, strong early praise on personality/writing, and enterprise demos citing fewer tokens and better task accuracy. Soul and Luna are sold on bringing Astra-class strengths into cheaper, faster models for scale — Altman's 'use tons of AI' line. Check primary pricing and Artificial Analysis pages before changing spend; several other episode numbers remain announcement-relayed.
Discourse winner ≠ stack winner
Buzz skewed to Opus. That does not retire Soul/Luna. OpenAI is filling price/modality slots for multi-model architectures; Anthropic is reclaiming Claude Code momentum. Soul as affordable daily driver and Opus for ambitious coding/visual work is a coherent split.
What pacing can look like
Four releases in two days is not a freeze — and may not be a frontier jump either. Iterating mid-tier classes with real cost cuts still opens agent use cases (Jevons on tokens). Map tasks to slots; upgrade where the job matches; skip the universal champion speech.
