summree
Fable 5 just returned & GPT 5.6 is coming (Huge Week in AI)
OpenAI
Riley Brown

Fable 5 just returned & GPT 5.6 is coming (Huge Week in AI)

⏱ 26 min video · 3 min read3 Jul 2026
TL;DR
Claude Opus 5 (called 'Fable 5' throughout) returned after a US government-mandated suspension, but with expensive API costs, rerouting issues, and a July 7 cutoff for plan-included access. The video also covers Anthropic's underwhelming Sonnet 5 release, the new Cursor iOS app, and a prediction that selling Slack-based AI agents will be the next big entrepreneurial opportunity.
Key points
1
Claude Opus 5 was re-released on approximately June 30, 2026 after being suspended 78 hours after its June 9 launch due to a US government national security directive; it now carries stricter safeguards and nationality verification requirements.
2
Plan-included access to Claude Opus 5 ends July 7, 2026 — after that it is API billing only, and costs are extreme: one 4-prompt coding session cost $174 and a single analysis prompt cost $135.
3
A known issue: Opus 5 reroutes mid-task to Opus 4.8 for tasks flagged as security risks; one streamer paid $321 for a session where only $78 went to Opus 5 and $242 to Opus 4.8.
4
Anthropic's Sonnet 5 launched Tuesday and is disappointing — despite costing less per token than Opus 4.8, it uses far more tokens per task, making it actually more expensive to run and roughly comparable to the open-source GLM 5.2 from Zhipu AI.
5
Cursor released an iOS app that lets developers fire off cloud agents from their phone — agents write code, test the app, screen-record results, and send back a PR — but it requires GitHub knowledge and is rated a 5/10 in technical difficulty.
Actionable insights
Do not measure AI model value by raw token price — always calculate cost per completed task; Sonnet 5 costs more per task than Opus 4.8 despite a lower token rate.
If you are on a paid Claude plan, use your Opus 5 plan-included access before July 7, 2026 — after that you pay full API rates which can reach $135+ per prompt for complex tasks.
To prepare for the agent-selling opportunity, Riley recommends studying: agent instruction files (system prompts), agent skills, external tool connections, safeguards, and which platforms (Slack, iMessage, voice) to deploy on — start narrow with one specific task before expanding.
Notable quotes

That game that I created was really good, but it costed me $174. $174. That is a lot of money. And this was literally four prompts.

If there is one takeaway from this section of the video, it is that don't always trust these raw token costs. What you need to look at is how much does it cost to complete a specific task.

Whoever allows people to create AI agents or templates that people can customize for specific workflows and allow people to just add them into Slack, that is going to be a very successful massive company.

Worth watching?
⏭️
Worth watching the full video?
The key facts, costs, and predictions are all captured here — skip the video unless you want to see the live game demo built with Claude Opus 5 or hear the Cursor iOS app walkthrough in motion.
Topics
AI & TechOpenAI

Explore more summaries on these topics →

Saved you some time? The creator still deserves a like.

Watch on YouTube →
More like this

Want this for your own channels?

Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.

Try it free