State of AI: Dec 2025
Curated list of the latest frontier models
GPT-5.2 (Thinking)
Just released. OpenAI's specialized model for deep scientific and mathematical reasoning. Sets new SOTA on FrontierMath and GPQA Diamond benchmarks.
View Release Notes →Claude Opus 4.5
Anthropic's new king of coding and agents. Features "Computer Use" capability (66.3% on OSWorld) and massive context window. Best for complex workflows.
Try Claude →Gemini 3 Pro
Google's native multimodal beast. Processes video, audio, and code in parallel. Known for "Vibe-Coding" capabilities and integration with Vertex AI.
Explore Gemini →Grok 4.1
xAI's latest model with direct access to X (Twitter) data. Scored highest on EQ-Bench3, making it the most emotionally intelligent model available.
Learn more →DeepSeek-V3.2
The new open-weight challenger. "Speciale" version matches proprietary models in logic tasks while being significantly more efficient to run.
View Github →GPT-5
The standard "super assistant" released in August. Unifies text, vision, and voice into a single model. The daily driver for most ChatGPT users.
Read System Card →Sora (v1.5)
Now widely available. Creates hyper-realistic video up to 2 minutes. Recently integrated with Disney IP for specific creative tools.
Watch Demos →Genie (Luma)
Generates fully textured, rigged 3D models in under 10 seconds. The industry standard for rapid game asset prototyping.
Create 3D →