← OpenAI
Matt Wolfe
AI News: Fable's Back But This New Model is Better?
⏱ 29 min video · 3 min read3 Jul 2026Worth watching
TL;DR
Matt Wolfe covers the return of Claude (referred to in-transcript as 'Fable') after a government-mandated shutdown, the announcement of GPT-5.6 models locked behind government approval, and Claude Sonnet 5's quiet launch. He also demos practical apps he built with Claude and introduces his 'BeautyBench' SVG benchmark for tracking AI model progress over time.
Key points
1
Claude (called 'Fable 5' in transcript) was shut down June 12 by the US government over security vulnerabilities found by Amazon, then relaunched July 1 with stricter safety guardrails that reportedly tanked coding benchmarks (debugging score dropped from 86.2 to 25.9 per Bridgemind's tests).
2
GPT-5.6 launched in three tiers (Soul, Terra, Luna) but is restricted to select government-approved partners; Soul Ultra benchmarks above Claude on TerminalBench (91.9% vs 84.3%) at roughly half the price of Claude.
3
Claude Sonnet 5 launched quietly at $2 input/$10 output (rising to $3/$15 after September 1, 2026) but is openly acknowledged by Anthropic as inferior to both Opus and Claude (Mythos class) models.
4
Google released Imagen 2 Light (4-second generation, ~$0.035 per 1,000 images) and Gemini Omni Flash for video editing/generation via API; NotebookLM now supports 60-second vertical short video generation.
5
Access to Claude on paid plans ends July 7, 2026, after which it requires additional usage credits on top of existing subscriptions.
Actionable insights
→
If you are on a paid Claude plan, use it heavily for large projects before July 7, 2026 — after that date you will need to pay extra usage credits to access it.
→
For API users, Claude Sonnet 5 at $2/$10 per million tokens (until September 1) offers a cost-effective middle ground if you do not need full Claude or Opus power.
→
OpenRouter can be used to bulk-generate outputs across many AI models simultaneously, as Matt demonstrated when building BeautyBench across dozens of models at scale.
→
Gemini Omni Flash is now available to developers via the API and Google AI Studio at $0.10 per second of video output — strong for editing existing footage, weaker for generation from scratch versus Sora or Veo.
→
The OpenAI proposal to give the US government a 5% equity stake (~$42.6B) in OpenAI raises significant regulatory conflict-of-interest concerns worth monitoring.
Notable quotes
“If the government has a stake in the company that the government is also in charge of trying to regulate, doesn't that create some pretty massive conflicts of interest?”
“Time you enjoy wasting is not wasted time.”
“You don't even have to get too deep into their system card for them to be going, yeah, it's a new model, but it's not as good as their other models.”
Worth watching?
✅
Worth watching the full video?
Watch if you want to see the live demos of the Claude-built apps and BeautyBench in action — the written summary covers all the news, but the visual walkthroughs of the SVG benchmark and short-form video dashboard are genuinely useful to see.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free