Easy AI Daily News Digest | December 5, 2025

A curated roundup of AI industry news for December 5, 2025, covering model releases, research, business moves, community discussions, and tool updates.

Model Releases & Updates

• Google Gemini 3 Deep Think: Available to Google AI Ultra subscribers, using parallel thinking to boost complex reasoning. Scores 45.1% on ARC-AGI-2, far exceeding GPT-5.1's 17.6%. Supports math, science, and other tasks. (Google AI, Google DeepMind)
• OpenAI GPT-5.1-Codex Max: Available via the Responses API and integrated into the Codex agent harness; works in IDEs like VS Code and Cursor. (OpenAI Devs, Cursor)
• Microsoft VibeVoice-Realtime-0.5B: Lightweight real-time text-to-speech model supporting English and Chinese, open-sourced on Hugging Face. (Model page)
• Nous Research Hermes 4.3: Built on ByteDance Seed 36B, approaching Hermes 4 70B performance; trained via the Psyche network with MoE support. (Blog)
• Mistral Large 3: Ranked #1 among open coding models on lmarena; available via Ollama cloud. (Mistral AI, Ollama)

Research & Technical Progress

• Google Titans architecture: Combines RNN efficiency with Transformer performance, supporting 2M+ token contexts; early results shown at NeurIPS. (Google Research)
• TorchAO MoE quantization: New MoEQuantConfig enables quantization of mixture-of-experts models for faster inference. (PyTorch PR)
• VATTENTION: First sparse attention mechanism with (ϵ, δ) guarantees, improving long-text processing. (arXiv)
• STRAW: Sample-tuned rank-augmented weights mimic neuromodulation, dynamically adjusting model weights for better task adaptability. (Substack)
• Fast ODE solver for diffusion models: Generates 4K images in 8 steps with quality comparable to 30-step DPM++2M SDE; open-sourced on Hugging Face. (Space, arXiv)

Industry & Funding

• Anthropic acquires Bun: Claude Code revenue reaches a $1 billion annual milestone. (Anthropic)
• Perplexity: Receives investment from football star Cristiano Ronaldo. (Tweet)
• Harvey: Raises $160M Series F at an $8 billion valuation, serving 700+ law firms. (Tweet)
• Antithesis: Raises $105M led by Jane Street for deterministic simulation testing of AI-generated code. (Tweet)

Community Discussions

• GPT-5.1 reportedly finds code bugs that Gemini 3 misses, per OpenAI Discord users. (Discord)
• Reddit users report the Z-Image model still filters gore/nudity despite claims of being uncensored. (Reddit)
• Debates on AI's impact on software jobs: users expect roles to change rather than disappear. (Reddit)
• LocalLLaMA users test VibeVoice-Realtime, praising English/Chinese support while noting Mandarin accent issues. (Reddit)
• Discussion of Gemini 3 Deep Think's ARC-AGI-2 score of 45.1% versus GPT-5.1's 17.6%. (Reddit)

Tools & Platform Updates

• OpenRouter "State of AI" report: Analyzes 100 trillion tokens; open models are 50% roleplay, paid models 50% coding, with Claude handling 60% of coding workloads. (Report)
• Windsurf: Integrates GPT-5.1-Codex Max with free trials for paid users and Low/Medium/High reasoning levels. (Announcement)
• mcp-apps-sdk: Open-sourced by General Intelligence Labs for embedding ChatGPT apps on other platforms. (GitHub)
• tinygrad: PR fixes train_step not utilizing input tensors, improving training efficiency. (PR)
• DSPy: Proposal to natively integrate Claude Code's Read/Write/Terminal tools. (Discord)

Benchmarks & Performance

• Gemini 3 Deep Think: 45.1% on ARC-AGI-2, a 2.5x improvement over GPT-5.1.
• Mistral Large 3: #1 in lmarena open coding model rankings.
• DeepSeek V3.2: Baseten serving metrics show 0.22s TTFT and 191 tps; strong lmarena rankings in math, law, and science. (Baseten)
• GPT-5.1-Codex Max: Positive user feedback on code quality and efficiency in Cursor and other IDEs.

*Source: Easy AI education project.*