Good Morning! Here's what I have for you in today's newsletter:
Anthropic introduces Fable 5.1 and Mythos 5.1, its most advanced models yet
Meta introduces Muse Voice Transcribe, its first real-time audio model
Google brings agentic video understanding to Gemini
Stop Fable 5.1 from burning through your weekly credit with these fixes
4 new AI tools worth trying today
AI MODELS
Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, described as the world's most advanced models for coding and knowledge work, with Fable 5.1 built for complex, long-running tasks and cache reads costing 75% less than Fable 5's.
Fable 5.1 is built for long, complex tasks, and its research skills give an early look at how AI could help with real scientific work.
It can match or beat Fable 5 at a much lower cost when you turn the effort level down, so you get to choose between speed and savings.
Cache reads now cost 75% less than before, which cuts total cost by about 25% for normal use and up to 45% for heavy agent work.
Cache pricing usually gets ignored next to the bigger capability claims, but it's often the real driver of cost once you're running the same context over and over. A 45% cut on agentic tasks matters most here, since those tasks repeat that same context again and again, so the savings add up instead of only happening once.
AI AGENTS
Meta introduced Muse Voice Transcribe, its first real-time audio perception model, delivering streaming ASR, diarization with 20+ speakers, and endpointing, ranking first on both Artificial Analysis's streaming speech-to-text benchmark and public diarization benchmarks.
It can tell apart more than 20 different speakers at once, which is far more than most transcription tools can handle.
It understands multiple languages and can follow a conversation even when people switch languages mid-sentence.
Accuracy gets even better when you feed it specific vocabulary, keywords, or context ahead of time.
Most speech models struggle once more than a few people talk at once, even the well-funded ones. Doing well on both tests at the same time suggests Meta built this to actually handle that problem, not just added it on afterward.
AI TOOLS
Google added agentic video understanding to its latest Gemini models, letting developers process long-form video content with more accuracy while using up to 88% fewer tokens.
This is built specifically for long videos, which normally use up a huge number of tokens to process.
Cutting token use by 88% means you can analyze a lot more video for the same cost.
Accuracy actually improved too, so this isn't a tradeoff, it's a real efficiency gain.
Video has always been expensive to process, since even a short clip turns into way more tokens than the same amount of text. Cutting that by 88% while still improving accuracy makes a real difference, turning something like analyzing a full lecture or meeting from too costly into actually doable.
HOW TO AI
Anthropic shipped a set of cost tools alongside Fable 5.1, and hardly anyone using it knows they exist. Four specific changes, confirmed straight from Anthropic's own team, can meaningfully cut what you're spending without touching output quality.
Step 1: Sweep your effort level
The single biggest lever. Effort has five levels, low to max, and higher effort means more tokens whether the task needs it or not. Start at high, the default. Drop to medium or low for routine tasks once quality holds.
Run this request at low effort and tell me what changed in the response
On CursorBench, Fable 5.1 at low effort scored higher than Fable 5 at high effort, at a third of the cost.
Step 2: Run cost-optimize
Profiles your own usage and ranks the cost levers that matter for your project.
/claude-api cost-optimizeYou approve or skip each suggestion, nothing changes without confirming first.
Step 3: Run prompt-audit
Prompts and skills accumulate cruft over time, verification rituals, emphasis boosters, stale scaffolds. Every one costs tokens on every request until removed.
/claude-api prompt-auditStep 4: Migrate old API configs
Handles the model ID swap plus breaking parameter changes across your codebase.
/claude-api migrate this project to claude-fable-5-1
P.S. You can access all the AI trainings (including the full version of this one), prompts and workflows if you upgrade.
Perplexity introduced hybrid compute in Perplexity Computer, letting a task start in the cloud and then move to a local model running on a Mac, built specifically for steps involving private files or sensitive data.
Google rolled out Google Pics to Workspace customers and Google AI Pro and Ultra subscribers, letting anyone edit individual objects, refine or translate text, and collaborate with a team on precise AI image creation.
OpenRouter launched Mercury 2.5 Preview from Inception AI exclusively on its platform, reaching 1,107 tokens per second through parallel token generation, with tunable reasoning and parallel tool calls built for latency-sensitive workloads.
🌍 World Labs: Atlas, the first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D, currently in early access.
⚖️ Harvey: Horizon Scanning monitors 12,000+ sources across 100+ jurisdictions, surfacing regulatory updates with impact and risk assessment built in.
🎵 Adobe Firefly: Generate Music, Speech, and Sound Effects are now generally available, commercially safe AI audio built directly into the creative studio.
🤖 AccuKnox: AgentZ brings agents, sandboxes, workflows, and governance into one platform to build, run, and control AI agents at scale.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial