KOLs Prefer Claude Opus 5.5's "Taste" Over Efficient GPT-6.1 Sol
In early October, several tech KOLs shared hands-on comparisons of GPT-6.1 Sol against Claude Opus 5.5 and Sonnet 5.5. The consensus: GPT-6.1 Sol is efficient and handles many tasks, but they personally prefer Anthropic's new models, saying the company has recovered Claude's distinctive "taste" while remaining fast and capable. Counterpoints emerged as well, with one user criticizing Opus 5.5 for adding unrequested creative changes. The debate suggests "taste" is becoming a new dimension for evaluating frontier models.
Confirmed
- kimmonismus stated GPT-6.1 Sol is efficient and suited to many tasks, but he now prefers Opus 5.5 and Sonnet 5.5, saying there's no going back after using Opus 5.5, and expressed anticipation for Anthropic's future releases.
- AiBreakfast agreed that Anthropic has recovered Claude's "taste" and that the new models are fast and efficient, half-jokingly calling for a dedicated benchmark for taste.
- These views were repeated by minchoi and other accounts, reflecting the same source being relayed.
- Academic Andrey Kurenkov called Opus 5.5 "the real deal," saying it's the first time in a while he's felt a significant leap in agentic coding capability; he noted he hadn't tried a certain other product.
- User VraserX reported the opposite experience: even with precise design requirements, Opus 5.5 insists on its own "creative" modifications, while GPT-6.1 Sol implements requirements strictly; he doesn't need an AI designer, just faithful execution.
Unconfirmed
- The superiority in "taste" is subjective; the posts include no quantitative benchmarks or test data.
- Specific capability differences (speed, task performance) lack measured evidence.
Why it matters
- "Taste" is emerging as a new evaluation dimension beyond efficiency and accuracy; AiBreakfast's call for a taste benchmark could shape future evaluation frameworks.
- Positive word of mouth from kimmonismus and Kurenkov's agentic-coding judgment point to a reputational rebound for Anthropic in the high-end model race, while VraserX's complaint reveals the flip side of "taste"—autonomous flourishes can be seen as interference by users who need faithful execution. The tension between these demands is worth watching.
2026-10-01 ~ 2026-10-03 · 7 related posts
Primary sources
- "GPT-6.1 Sol Is Efficient, But Claude Opus 5.5 Has Regained Its Taste": KOL Weighs In — kimmonismus ·
- User Gripes Claude Opus 5.5 Adds Unwanted Creative Twists While GPT-6.1 Sol Just Follows the Brief — VraserX ·
- Opus 5.5 is the real deal: first model in a while to meaningfully up agentic coding, says researcher — andrey_kurenkov ·
- [source] "GPT-6.1 Sol Is Efficient, But Claude Opus 5.5 Has Regained Its Taste": KOL Weighs In — kimmonismus · 2026-10-01
- 'Can't go back': AI KOL hails Anthropic's Opus 5.5 as a step change — kimmonismus · 2026-10-01
- Claude's 'taste' beats GPT-6.1 Sol, sparking calls for a taste benchmark — AiBreakfast · 2026-10-02
- [source] User Gripes Claude Opus 5.5 Adds Unwanted Creative Twists While GPT-6.1 Sol Just Follows the Brief — VraserX · 2026-10-03
- [source] Opus 5.5 is the real deal: first model in a while to meaningfully up agentic coding, says researcher — andrey_kurenkov · 2026-10-03
2 near-duplicate retellings: minchoi · kimmonismus