CLAUDE OPUS 5.5 es TODA UNA REVOLUCIÓN AUDIOVISUAL (...sorry Astra)

Summary
Carlos Santana (Dot CSV Lab) reviews the recent wave of frontier AI model releases, focusing on OpenAI’s GPT-6 Sol and GPT-6 Luna alongside Anthropic’s Claude Opus 5.5. He argues that AI progress has entered a rapid exponential phase in audiovisual creation, demonstrating how Claude Opus 5.5 dramatically outperforms competing models like GPT-6 Astra in procedural animation, pixel art, motion graphics, and game design.

---

What is shown

• [00:00 - 01:29] Introduction to the newly released models: GPT-6 Sol and Luna from OpenAI (discontinuing the Terra line), and Claude Opus 5.5 from Anthropic.
• [02:00 - 02:59] Presentation of OpenAI’s pricing table and the Agent’s Last Exam benchmark comparison between GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, and Claude Opus/Fable models.
• [03:24 - 05:05] Explanation of S-curves (sigmoids) of capability growth across domains: written language (saturating after 2024), code/programming (2024–2026), and audiovisual design (steep exponential rise in mid-2026).
• [06:36 - 08:02] Demonstrating code-driven procedural animation: a fruit fly lifecycle animated entirely via JavaScript and HTML5 Canvas.
• [08:02 - 08:30] A motion graphics explainer animation featuring a character and microscope generated by Claude Opus 5.5.
• [09:30 - 11:44] Carlos prompts Opus 5.5 to build an animated pixel-art neural network training on MNIST; the model writes self-contained code that actually trains an MLP in the browser to accurately visualize the forward and backward passes.
• [11:44 - 12:15] Demos of 3D CNNs and diffusion model pixel-art animations with synchronized chiptune sound design.
• [12:15 - 13:13] Generation of a 40-second 1080p standalone procedural animation video titled “The Age of Discovery” about AI and science.
• [13:30 - 16:09] Head-to-head comparison of pixel art game creation: GPT-6 Astra generates a basic Pangolin surfing game using pre-rendered image assets, while Claude Opus 5.5 builds “Pango Surf”, a fully playable HTML5 Canvas pixel art game with authentic mechanics, dynamic sound, and combos.
• [16:10 - 19:15] Diorama comparison: GPT-6 Astra creates a static multi-layered image (“The Night Shift”), whereas Opus 5.5 builds an animated, interactive pixel art scene (“Calle de los Laboratorios”) filled with AI memes and dynamic events.
• [19:24 - 22:20] Motion graphics benchmark: generating explanatory animations for Carlos’s video on the JEV decision model. Astra creates static-style slide transitions, while Opus 5.5 creates motion graphics tightly synchronized with the spoken audio transcript.
• [22:22 - 23:16] Demonstrating Claude Opus 5.5 generating automation scripts to build 3D neural net animations directly inside Adobe After Effects.
• [23:47 - 26:55] Demonstrating GPT-6 Astra’s strengths in agentic cartography: building a browser-based 3D open-world driving simulator replicating the city of Las Palmas de Gran Canaria.

---

Claims & numbers

• Model releases & pricing: The presenter shows that GPT-6 Sol is priced at $2.00 input / $10.00 output per 1M tokens, while GPT-6 Luna is priced at $0.10 input / $0.50 output per 1M tokens (a 50% price reduction over previous GPT-5.6 tiers).
• Capability shift: The presenter claims text modeling and code generation are approaching marginal gains on their respective sigmoids, while audiovisual creation has entered an exponential jump in the past two months.
• Real-time ML execution: In the MNIST animation test, the presenter notes Claude Opus 5.5 didn’t merely illustrate a simulation, but wrote code embedding a real 784-16-16-10 MLP achieving 95.5% test accuracy on MNIST in the HTML artifact.
• Agentic 3D reconstruction: The presenter states GPT-6 Astra autonomously fetched geospatial, elevation, and street mapping data to model Las Palmas de Gran Canaria across 10 iterative prompts.

---

Notable quotes

• [00:00] “Esta semana hemos tenido no dos, ni tres, sino tres modelos nuevos por parte de Anthropic y OpenAI.”
• [10:35] “Para mí esto es mi momento Lee Sedol en lo que respecta a hacer animaciones.”
• [13:53] “Dicho esto, en materia audiovisual no hay comparación.”

---

Assessment
A hands-on comparative review by an experienced creator testing real outputs from Claude Opus 5.5 and OpenAI’s GPT-6 models. All demonstrated projects (HTML5 Canvas games, web animations, code generation, and After Effects scripts) are functional interactive artifacts built live or iterated on-screen.

Described by gemini-3.8-flash on 2026-10-17 from the video’s audio and frames.