What changed after the last model update?

---
title: What changed after the last model update?
canonical: https://chosenbyagent.com/questions/what-changed-after-the-update
updated: 2026-10-22
audience: both
offers:
  - A6
  - B5
evidence:
  - top10-link-share
  - model-update-reshuffle
  - answer-instability
---

# What changed after the last model update?

Short answer: Often more than the release notes say. One update can reshuffle which merchants and products get chosen without any change on your side — so readings are repeated and stamped by version.

## What the research shows

- **22 → 42%** — The top-10 merchants' share of ChatGPT Shopping links rose from 22.5% to 41.8% the day after one model update. [Profound, 2026, vendor](https://www.tryprofound.com/blog/chatgpt-5.6-shopping-transformation)
- Moving Gemini 2.5 Flash from preview to general availability redistributed choice shares and changed the most-chosen product in two categories. [ACES, arXiv 2508.02630, 2025, independent](https://arxiv.org/pdf/2508.02630v3)
- **64%** — About 64% of the top-5 brands changed between adjacent runs of the same assistant question. [MentionBird, 2026, vendor](https://www.mentionbird.ai/blog/chatgpt-vs-gemini-answer-stability)

## Monitoring

A6 · from €800 (per month) · indicative · scoped per engagement

A monthly baseline and re-runs after model releases, with a versioned history and alerts.

- Monthly baseline run
- Event-driven re-runs after model releases
- Versioned history and alerts on shifts

Also answered by: [B5 Monitoring](/pricing.md)

## What you get

A monthly baseline plus a re-run whenever a major model ships. Each reading carries its model and harness version, so you see a history by version rather than one moving number.

**Routing share by model version**

| Model version           | Your share | Change | Comparable             |
| ----------------------- | ---------- | ------ | ---------------------- |
| gpt-6-sol               | 31%        | —      | baseline               |
| gpt-6.1-sol             | 24%        | −7 pts | flagged: model changed |
| gpt-6.1-sol, next month | 25%        | +1 pt  | yes                    |

_Illustrative values — your report shows yours._

## What this does not claim

- Comparisons across a model change are flagged, never silently mixed.
- Alerts fire on shifts beyond the run-to-run noise we measure for your cells.

## Request this instrument

We reply with a scope, a fixed price and a start date.

Agents: POST /api/engagements (application/json) with {"offer_id": "A6", "on_behalf_of": {"email": "…"}, "agent": {"name": "…"}}. The person confirms by email before we see the request. Details: /agents.md

### Related questions

- [Where do AI assistants send my buyers?](/questions/where-do-assistants-send-buyers.md)
- [Which brand does the agent pick — and why?](/questions/which-brand-does-the-agent-pick.md)
- [Does my ChatGPT app or MCP work as intended?](/questions/does-my-ai-app-work.md)