---
title: "Inline citations, call summaries and new AI models"
canonical_url: "https://chatthing.ai/blog/citations-call-summaries-and-eight-new-models"
last_updated: "2026-10-14T16:14:27.055Z"
meta:
description: "See which page backs each answer with inline citations, get a summary by email or text after every call, and try GPT-6, Claude Haiku 5.5 and six more models."
"og:description": "See which page backs each answer with inline citations, get a summary by email or text after every call, and try GPT-6, Claude Haiku 5.5 and six more models."
"og:title": "Inline citations, phone call summaries and eight new models - Blog - Chat Thing"
"twitter:description": "See which page backs each answer with inline citations, get a summary by email or text after every call, and try GPT-6, Claude Haiku 5.5 and six more models."
"twitter:title": "Inline citations, phone call summaries and eight new models - Blog - Chat Thing"
---
[← Back to blog](https://chatthing.ai/blog)**Changelog**# **Inline citations, phone call summaries and eight new models**
~ 5 min read
---
---
### **On this page**
1. [Sources that match the answer](#sources-that-match-the-answer)
2. [Your phone agent now reports back](#your-phone-agent-now-reports-back)
1. [A summary after every call](#a-summary-after-every-call)
2. [Keep the number you already have](#keep-the-number-you-already-have)
3. [Ring your phone first](#ring-your-phone-first)
4. [Send every call to your other tools](#send-every-call-to-your-other-tools)
3. [Eight new models](#eight-new-models)
1. Models we're retiring
4. [Smaller improvements](#smaller-improvements)
5. [Try it](#try-it)
Today's Chat Thing release covers three things. Your agent now shows exactly which pages an answer came from. Your phone agent tells you who called and what they need. And there are eight new models to try, from GPT-6 to Claude Haiku 5.5.
## Sources that match the answer
With **Include sources for responses** turned on, every reply used to list every page your agent retrieved, whether the answer used it or not. A question about pricing could end with links to your returns policy and your about page. That makes the answer harder to check, and it makes the sources easy to ignore.
Now Chat Thing checks which pages the finished reply actually used and lists only those. Each source is numbered, and the same numbers appear in the reply next to the part each page supports. Hover over a number to see the page title.
The numbers appear once per paragraph or list, so they don't clutter every sentence. The sources load a moment after the reply finishes.
A few details:
- It works in your website chat. Slack, WhatsApp, email and phone replies are unchanged.
- Only pages from website data sources are listed.
- It's available on every plan. Turn on **Include sources for responses** on your agent's **General settings** tab.
- If you build on the [Chat API](https://chatthing.ai/docs/developers/chat-api), `sources` now lists only the cited pages, with offsets showing where each marker belongs.
See [Retrieval settings](https://chatthing.ai/docs/knowledge/retrieval-settings) for how sources fit with your other knowledge settings.
## Your phone agent now reports back
An agent that answers your phone is only useful if you find out what happened. If you're up a ladder, with a client or driving, you shouldn't have to read every transcript to know who needs a call back.
### A summary after every call
A few seconds after each call, Chat Thing summarises it and tells you who called, why, how urgent it is, the details they gave (such as a postcode or a preferred time) and what your agent told them would happen next.
- **Email** is on by default. Summaries go to up to three addresses, or to the team owner.
- **Text** goes to up to three UK mobiles, for every call or only urgent ones. Texts come from your Chat Thing number, so this needs a UK number. A privacy option sends just who called and a link, for businesses that don't want details on a lock screen.
- **You decide what's urgent.** Describe it in plain words, such as "burst pipes, no heating" or "severe pain or swelling".
- **Spam, sales calls and wrong numbers** are summarised but never emailed or texted.
- **Hang-ups** can send an alert too, because a caller who hung up before speaking is often worth ringing back.
Click **Send a test summary** to see exactly what arrives. The call log now shows why each person called, with urgent calls flagged.
### Keep the number you already have
You don't have to print a new number. Keep the one on your van, website and Google listing, and forward the calls you miss to your Chat Thing number.
Choose **Keep my number**, pick your phone network, and Chat Thing shows the divert code to dial, the code to undo it and tips for that network. Then click **Test forwarding**. Chat Thing rings your business number to check the call really forwards, and tells you what to change if it doesn't.
### Ring your phone first
Prefer to answer yourself when you can? Choose **Ring me first**. Calls to your Chat Thing number ring your phone for 10 to 30 seconds. You hear who's calling and press any key to take it. If you don't, your agent answers.
Time you or your team spend on a call, whether you picked up a ring-first call or took a transfer, now uses voice minutes at half rate.
### Send every call to your other tools
The new **Call ended** webhook sends each call's summary to your CRM, job management or booking software: caller name, reason, urgency, call type, the details they gave and the follow-up.
The Phone channel is available on Standard, Pro and Enterprise with a voice add-on, with numbers in the UK and US. Everything above is in [Answer phone calls with your agent](https://chatthing.ai/docs/channels/phone).
## Eight new models
Eight models join [Chat Thing's model line-up](https://chatthing.ai/models). All of them support vision and Power-Ups.
| Model | Provider | Context window | Message tokens |
| --- | --- | --- | --- |
| GPT-6 Luna | OpenAI | 1,050,000 | Light |
| GPT-6 Sol | OpenAI | 1,050,000 | Heavy |
| GPT-6.1 Sol | OpenAI | 1,050,000 | Heavy |
| Claude Haiku 5.5 | Anthropic | 1,000,000 | Light |
| DeepSeek V4.1 Flash | DeepSeek | 1,048,576 | Light |
| MiMo V2.6 Flash | Xiaomi | 1,050,000 | Light |
| MiniMax M3 | MiniMax | 1,048,576 | Medium |
| Command A+ | Cohere | 192,000 | Medium |
MiniMax and Xiaomi are new providers in Chat Thing.
For a busy support agent, start with the light models: GPT-6 Luna and Claude Haiku 5.5 are the ones we'd test first. GPT-6 Sol and GPT-6.1 Sol use far more message tokens, so save them for harder, multi-step work. GPT-6.1 Sol always does some reasoning before it answers.
Whichever you try, test it with your own questions before you switch a live agent. Check the answers, the tone and whether it calls the right Power-Ups. Choosing a model is available on Standard, Pro and Enterprise.
### Models we're retiring
Nine models are leaving Chat Thing:
- **Gemini 2.5 Pro, Gemini 2.5 Flash and Gemini 2.5 Flash Lite** retire on **20 October 2026**, when OpenRouter stops serving them.
- **GPT-5.1 Chat, Claude Haiku 3, Claude Opus 4, Claude Opus 4.7 (Fast), Claude Opus 4.8 (Fast) and Claude Opus 5 (Fast)** have already been removed by their providers.
If one of your agents uses a Gemini 2.5 model, switch it to a current model before 20 October. We'll move any agent still on a retired model to the closest current model from the same provider.
## Smaller improvements
- **What's new** in your dashboard now has an unread badge and follows your light or dark theme.
- The assistant that helps you set up a new agent now runs on Claude Sonnet 5.5, and the avatars it creates use GPT Image 2.5 Flare.
- The team switcher works again after your first switch. It had stopped opening.
- Voice calls to our own help agent failed with "Team not found" if you were signed in. That's fixed.
## Try it
Turn on **Include sources for responses** for your website agent, then ask it a question your content answers and check the numbered sources. If you use the Phone channel, open **After each call** and click **Send a test summary**.
### **Related articles **
[](https://chatthing.ai/blog/voice-calls-and-phone-numbers) [Chat Thing agents can talk: voice calls and phone numbers are here
](https://chatthing.ai/blog/voice-calls-and-phone-numbers)
Chat Thing agents can now hold voice conversations in your web chat and answer a US or UK phone number, with the same knowledge, power-ups and AI model you've already set up.
[](https://chatthing.ai/blog/glm-5-3-flash-supportbench-and-link-controls) [GLM 5.3 Flash lands in Chat Thing, plus SupportBench and link controls
](https://chatthing.ai/blog/glm-5-3-flash-supportbench-and-link-controls)
GLM 5.3 Flash is now available in Chat Thing, alongside SupportBench model testing and control over where links in embedded agents open.
[](https://chatthing.ai/blog/gemini-3-7-flash-and-grok-4-6) [Gemini 3.7 Flash and Grok 4.6 land in Chat Thing
](https://chatthing.ai/blog/gemini-3-7-flash-and-grok-4-6)
Gemini 3.7 Flash and Grok 4.6 are now available in Chat Thing, alongside clearer Power-Up activity and support for more Google Drive file types.
- **Previous **
- [← Chat Thing agents can talk: voice calls and phone numbers are here](https://chatthing.ai/blog/voice-calls-and-phone-numbers)