Fable 5 Returns, Claude Sonnet 5 Lands, and Gemini Drops Two New Media Models

Anthropic's Fable 5 is back for paid subscribers with a July 7 deadline and benchmarks claiming double the remote work capability of Opus 4.8. Claude Sonnet 5 also quietly launched as the new default model, while Google released two new Gemini media models targeting fast, cheap image and video generation.

Fable 5 Is Back — With a Clock Ticking

Fable 5 has been re-released to all paid Anthropic subscribers, but with a tight window: it's only bundled into subscription plans until July 7, and users are capped at 50% of their standard limits.

According to Anthropic, this re-release includes stronger guardrails than the original launch. A new benchmark claims Fable can complete 16% of remote work projects — reportedly double the rate of Opus 4.8.

Theo has quick tips on getting the max out of Fable in the next 5–6 days.

If you're planning to use it, the window is short and the cap is real — plan accordingly.

Claude Sonnet 5: Capable but Costly in Practice

Claude Sonnet 5 launched alongside Fable's return, positioning itself close to Opus 4.8 on most agent benchmarks at a lower per-token rate: $2/$10 per million tokens with launch pricing valid until August 31.

It's now the default model for Free and Pro tiers, and is available in Claude Code and the API. However, early user reports — including the author's own — suggest it runs expensive and slow per task in practice, making it a tough sell over competing models at similar real-world cost.

Gemini Drops Two New Media Models

Google released a pair of new media-focused models:

Nano Banana 2 Lite — fast image generation (under 4 seconds), priced at roughly $0.033 per image at 1K resolution
Gemini Omni Flash — video generation and editing at $0.10 per second

Both are available in the Gemini app and via API today.

Notable Releases from the Ecosystem

A few other launches worth tracking this week:

Bridgewater and Thinking Machines co-trained a specialist financial model hitting 84.7% accuracy on financial triage at 13.8x lower cost than the best frontier model tested
Factory's Droid Shield 2.0 fine-tuned detectors to catch exposed secrets in agent sessions and reduce false positives
xAI launched a no-code Grok Voice Agent Builder at $0.05/minute
Ramp PorTAL enables moving fine-tuned tasks between models at roughly half the standard cost
Claude Science entered beta — a Mac/Linux workbench for scientists with 60+ scientific skills and connectors, available to Pro, Max, Team, and Enterprise users