Model change gpt-5-5 → gpt-5-6 on 2 prompts

Evidence

Before

gpt-5-5

After

gpt-5-6

Breadth

2/33 prompts · local

Modes

instant

Measuring days

22 Aug 2026 · 4 field diffs in events.jsonl

{
  "routing.model_slug": "gpt-5-6"
}

Daily data of 22 Aug 2026 →

Context

model/gpt-5-6 model/gpt-5-5

Impact

For tools: model_slug in the send payload and in routing.*. Standard mode gpt-5-6, thinking gpt-5-6-thinking. In the UI „GPT-5.6 Sol". A change of the slug is a baseline break of the time series (see version history).

For visibility: Model generations differ in when they search, how they phrase queries and which format web.run uses. A model change can shift visibility without anything changing on a website.

Status history

Open

On 21 Aug two prompts still answered with gpt-5-5 (K4-02, K8-01), from 22 Aug consistently gpt-5-6. Rollout in waves or chance?

How do gpt-5-6 and gpt-5-6-thinking differ in fan-out? Observation: thinking bundles more queries on product questions (up to 7) but barely thinks longer (2–6 s).