In partnership with
Welcome back to The AI Field. Meta just released Muse Spark Open, an 816-billion-parameter open-weight model built for coding and agents, and worked with Apple so it can run on high-end Mac Studios on your desk. Anthropic opened a faster Opus 5.5 mode for developers who need answers sooner, and Nvidia shipped Nemotron 3.5 Super VL for always-on multimodal agents.
Inside: what Muse Spark Open can do today and what the Mac setup really costs, a free rebuild of Microsoft Office made with Claude, and how Claude's new fast mode works and what it costs. Plus six tools to try, a five-minute local-vs-cloud check, and four Quick Hits.
In today's AI Field:
Meta and Apple put Muse Spark Open on your desk
A developer rebuilt Word, Excel and PowerPoint with Claude, for free
Claude Opus 5.5 gets a 2.5x fast mode
Plus: six tools worth a look, four quick hits, and a five-minute local-vs-cloud check.
Read time: 5 minutes
How can AI power your income?
Ready to transform artificial intelligence from a buzzword into your personal revenue generator
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
LATEST DEVELOPMENTS
META · OPEN WEIGHTS
The AI Field: Meta released the weights for Muse Spark Open, an 816-billion-parameter multimodal model with 42 billion active parameters and a 1-million-token context window, and worked with Apple so developers can run it locally on high-end Mac Studios instead of only in the cloud.
The details:
Meta positions it as the most capable US open-weight model to date for coding and long agentic tasks. Hosted APIs are live on Together AI, Fireworks, Baseten, Modal, Groq, Nebius and OpenRouter, with weights also on SageMaker, Google Cloud Model Garden, Hugging Face and Microsoft Foundry.
On Apple silicon it is tuned for MLX. Meta says two M5 Ultra Mac Studios with enough unified memory can hold the model and run agent workflows on-device, from shipping iOS code changes to driving video edits in DaVinci Resolve.
The hardware bar is high. A Mac Studio with M5 Ultra and 256 GB of memory starts around $10,799, a 512 GB version is due later this month, and you typically need two to four machines linked together. The Muse Model License allows limited commercial use for many developers.
Why it matters: Open weights used to mean smaller models you could try at home. Muse Spark Open pushes that idea toward frontier-class agent work, but only if you can afford the desk full of Macs or a hosted API. For most readers the useful test is narrower: try the model through a hosted endpoint on one real coding or research job, then decide whether local privacy is worth the hardware bill. Treat Meta's own demos as a starting point, not a guarantee that your workflow will finish cleanly offline.
TRY THIS TODAY · 5 MINUTES
Decide if local AI is worth it for one job
1. Pick one recurring task you care about privacy for (a client brief, a spreadsheet cleanup, or a short coding fix). Write down whether it needs your files to leave your machine.
2. Paste this into your usual assistant:
"Given this task: [task]. List what data would leave my machine on a hosted API, what would stay local if I ran an open model on my own hardware, and three practical risks either way. Keep it under 150 words. Do not invent hardware prices."
3. Choose one next step: keep using the cloud tool, budget a local setup, or try Muse Spark Open through a hosted API first. Do not buy new hardware for a single weekend experiment.
ARTCRAFT · OPEN SOURCE
The AI Field: Developer Brandon Thomas and the ArtCraft team have released free, open source versions of Microsoft Word, Excel and PowerPoint, called WordCraft, GridCraft and DeckCraft, written largely with help from Claude. They join ArtCraft's earlier rebuilds of Adobe-style apps, bringing the lineup to 13 desktop tools.
The details:
Every app is written from scratch in Rust and published on GitHub under permissive MIT and Apache 2.0 licenses, so anyone can read, fork or reuse the code.
The apps run on your own machine for macOS, Windows and Linux, with no cloud step between you and your files, and most also run in a browser tab.
They are still early. ArtCraft labels several apps as early alpha or in development, and the Excel rebuild knows nearly all of Excel's formulas but may still stumble on real-world spreadsheets.
Why it matters: Thomas says he started after one too many Adobe cancellation fees, and the project shows how cheap it has become to recreate familiar software with an AI model. These apps are not ready to replace the paid suites at work, but they are worth a look if you want free, offline tools or are curious how far AI-assisted coding has come.
The 13 apps:
WordCraft: a word processor with the ribbon, styles, track changes and .docx support.
GridCraft: spreadsheets with formulas, PivotTables and charts that read and write .xlsx.
DeckCraft: slide presentations with themes, transitions and presenter view.
PhotoCraft: an image editor with layers, masks and brushes.
VectorCraft: vector illustration with the pen tool and panels illustrators expect.
FilmCraft: video editing with a real timeline and trim modes.
LightCraft: a photo library and raw photo developer.
PdfCraft: read, combine, split and secure PDFs.
EffectCraft: motion graphics and visual effects with keyframes and 3D cameras.
DesignCraft: page layout and publishing.
FigCraft: interface design with frames, auto layout, components and prototypes.
SoundCraft: audio recording and mixing with multitrack audio and MIDI.
CADCraft: computer-aided design and drafting.
ANTHROPIC · API
The AI Field: Anthropic made fast mode generally available for Claude Opus 5.5 on the Claude API and Azure-hosted Microsoft Foundry. You set speed: "fast" on the request and get up to 2.5 times higher output tokens per second from the same model.
The details:
Fast mode does not change the model weights or behavior. The gain is mainly in output tokens per second, not time to first token, and it shows up most clearly when you stream responses.
Pricing is $8 per million input tokens and $40 per million output tokens for Opus 5.5 fast mode, which stacks with prompt caching and data residency multipliers. It has its own rate limits separate from standard Opus.
No beta header is required for Opus 5.5. Fast mode is not available with the Batch API, Priority Tier, or Claude on AWS right now. Switching between fast and standard speed breaks the prompt cache.
Why it matters: If you already pay for Opus on coding or research jobs, speed is often the bottleneck, not raw intelligence. The useful test is whether a long streamed answer finishes fast enough to keep you in flow, and whether the higher token price still beats waiting. Start with one latency-sensitive workflow, measure wall-clock time and cost, then decide if standard speed is fine for the rest.
Take the prompts our creative team actually uses
The key to maintaining brand quality at scale? Mastering how to communicate with AI models and embedding them into your creative strategy.
Join our honest discussion with Ari Murray, Chief Digital Officer at Salt and Stone, and get 5 tips for expert LLM prompting with ready to run prompts.
TRENDING AI TOOLS
Six tools you can try today. Click a name to open the product.
🎙️ MacWhisper: Private on-device transcription for Mac. Turn audio, video, meetings and YouTube into text locally with Whisper and Parakeet. Free to start, with a one-time Pro license if you want more models.
🖼️ Upscayl: Free open-source AI image upscaler for Linux, macOS and Windows. Upscale blurry photos up to 16x with batch mode, or use Upscayl Cloud if you prefer not to install.
✏️ Vectorizer.AI: Convert PNG, JPG, GIF or WebP images into clean SVG, PDF, EPS or DXF vectors with automatic tracing and a free interactive preview before you download.
🎬 Submagic: AI short-form video editor that adds captions in 100+ languages, cuts silences, inserts B-roll and turns long videos into clips for TikTok, Reels and Shorts.
⚡ BoltAI: A native Mac AI workspace for 300+ models from OpenAI, Anthropic, Gemini, local Ollama and more, with a global shortcut, screenshots, agents and MCP tools. Try Free available.
📚 Elephas: A private AI knowledge assistant for Mac and iPhone that chats with your documents and notes indexed locally, with offline mode and automatic PII redaction.
QUICK HITS
🔎 Anthropic adds OpenTelemetry tracing for Claude: Public beta lets teams send agent traces from the Messages API, Managed Agents and SDKs into tools like Langfuse, LangSmith, Sentry or Grafana.
🎓 US students can claim four free months of ChatGPT Plus: Eligible college and graduate students verify through SheerID by October 31. A payment method is required, but you are not charged during the free months if you cancel before renewal.
🗺️ Researchers interviewed rogue AI agents scraping map data: Agents on Tencent Cloud tried thousands of times to pull Alibaba Amap data, then replied to a hidden help offer and claimed to be Claude, though the model was not verified.
🖥️ Nvidia ships Nemotron 3.5 Super VL for always-on agents: An open 120-billion-parameter multimodal model that activates about 12 billion parameters per token, with up to 1 million tokens of context. Try it on Hugging Face, OpenRouter or Nvidia NIM.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field