Happy Sunday! We just had another crazy week in AI. Meta has released Muse Spark 1.1, OpenAI rolled out GPT 5.6, and there is a new GitHub repository with permanently free LLM APIs.
And that's not all, here are the most important AI moves you need to know this week.
NVIDIA just open-sourced its Skills repo, a catalog of 110+ verified "agent skills," portable instruction sets that teach AI agents how to use CUDA-X libraries and platform tools correctly. Instead of your coding agent guessing its way through cuOpt or NeMo docs, it loads a battle-tested playbook written by the product teams themselves.
One-command install: Install any skill in one line with
npx skills add nvidia/skills. The CLI prompts you to pick a skill and destination, so there's no cloning or copying folders by hand.Works everywhere: Skills work with Claude Code, Codex, Cursor, and Kiro out of the box, since they're built on the open agentskills.io spec, meaning the same SKILL.md runs reliably across agents.
Verified, not vibes: Every skill is signed with a signature verifiable against NVIDIA's trust anchor, and each one is scanned for hidden instructions, prompt injection, and supply-chain risks before publication.
Try it now → https://github.com/NVIDIA/skills
Meta just shipped Muse Spark 1.1, and it's a big deal because it's the first Muse model available through a public developer API. It's built for agentic work: writing and debugging code, using software tools, reviewing screenshots, and operating within software environments with minimal human oversight. And you don't pay a cent to start.
Free to try, two ways: Every new API account gets $20 in free credits, and it's free to use in Thinking mode inside the Meta AI app and on meta.ai.
1M-token context that manages itself: The context window is 1,000,000 tokens, and the model actively compacts it, dropping stale detail while keeping what a long-running workflow needs.
It sees screens and acts: Computer use decides on its own when to write a script versus click through a UI, with support for images, video, and documents in a single call. On JobBench (professional tool use) it scores 54.7 against Opus 4.8's 48.4 and GPT-5.5's 38.3.
Try it now → https://www.meta.ai
xAI just released Grok 4.5, its smartest model built for coding, agentic tasks, and knowledge work, and its strongest ever, trained alongside Cursor. The headline isn't a benchmark score. It's efficiency: this thing does the same work with a fraction of the tokens.
2x token efficiency: It achieves roughly 2x the token efficiency of comparable leading models, solving tasks in under half the number of steps. On SWE Bench Pro it averages 15,954 output tokens per task, about 4.2× fewer than Opus 4.8's 67,020.
Free right now: xAI is offering free Grok 4.5 usage for a limited time in Grok Build and Cursor, and Cursor is doubling Grok 4.5 usage allowances for the first week after launch.
Dirt cheap after that: Priced at $2 per million input tokens and $6 per million output, served at fast-model speeds of 80 TPS, well under Opus 4.8's $5/$25.
Try it now → https://grok.com/
ByteDance just released Seedream 5.0 Pro, and it kills the most annoying thing about AI image generation: the full re-roll. It reads the grounding, meaning where each element sits in the frame, so you can lock onto one target and change just that part while the rest stays exactly as it was.
Point-and-edit precision: Use point selection, lasso, box selection, sketches, color swatches, or material references to edit a specific region rather than regenerating the entire image.
Real editable layers: Every generated image can be broken into a background layer plus individual element layers, each exported as a transparent PNG, with up to 20 separated layers per request. Drag, scale, and recompose like a design file.
Flawless text in 14 languages: It renders correct letterforms and local typography across 14 languages including Arabic, Japanese, Korean, and Thai, with proper right-to-left flow for Arabic and accurate tone-mark stacking for Thai.
Try it now → https://seed.bytedance.com/en/seedream5_0_pro
OpenAI just made GPT-5.6 generally available, and it's a whole family: Sol is the flagship, Terra is the balanced everyday tier, and Luna is the most cost-efficient. The number identifies the generation, while Sol, Terra, and Luna are durable capability tiers that advance on their own cadence. The claim that matters: state-of-the-art coding at less than half the tokens.
New coding SOTA, half the tokens: On the Artificial Analysis Coding Agent Index, GPT-5.6 Sol sets a new state of the art at 80.0, which is 2.8 points above Claude Fable 5, while using less than half the output tokens and costing about one-third less.
Terra halves your bill: Terra is a balanced everyday model with GPT-5.5-competitive performance at 2x lower cost, at $2.50 input and $15 output per million tokens, with Luna at $1 and $6.
Ultra mode = built-in agent teams: The new ultra setting coordinates multiple agents across parallel workstreams, running four agents by default and lifting Terminal-Bench 2.1 from 88.8% to 91.9%.
Try it now → https://chatgpt.com
Awesome Free LLM APIs is a curated list of permanently free LLM APIs, with rate limits, OpenAI SDK compatibility, available SDKs, speed tiers, and free model lists. No trial credits, no time-limited promos, no credit card required. If you're prototyping, this is your new homepage.
Everything in one place: It covers first-party APIs from model creators like Google (Gemini's free tier is among the most generous anywhere), Mistral, Cohere, and Zhipu, plus inference platforms including Groq, Cerebras, OpenRouter, GitHub Models, NVIDIA NIM, Hugging Face, and Cloudflare.
Ready-to-paste code: The repo ships a provider dictionary with base URLs and model names for every service, and all OpenAI-compatible free APIs work with the same pattern. Swap the URL and key, keep your code.
Stack them for real capacity: Since each provider has independent rate limits, routing across Gemini + Groq + OpenRouter + Cerebras multiplies your free capacity, enough for genuine projects rather than just demos.
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