Daily AI Updates: December 5, 2025

Main Content

• GPT-5.1-Codex-Max Model Release: OpenAI has launched its most advanced agent-style coding model, supporting complex software engineering tasks and code generation. It is integrated into the Responses API, offering enhanced task decomposition and execution capabilities.

• GPT-5.1-Codex-Max Feature Update: The model is now in public preview in GitHub Copilot Pro/Business/Enterprise, improving performance on complex refactoring tasks and offering better stability for large codebases.

• GPT-5.1-Codex-Max Pricing Adjustment: OpenAI has set the API pricing at $1.25/million tokens for input and $10/million tokens for output, with direct availability on platforms like Cline.

• K2-V2 Model Release: MBZUAI has released a 70B parameter, 360-open LLM as a base model for inference adaptation, supporting a native 512K context window expandable via RoPE.

• K2-V2 Performance Optimization: MBZUAI reports 69.3% on the GPQA-Diamond benchmark (after SFT) and 62.1% on ArenaHard V2, surpassing Qwen2.5-72B and approaching Qwen3-235B.

• K2-V2 Feature Update: MBZUAI has added a mid-training inference stage and strong tool use scaffolding, open-sourcing over 250M inference trajectories, full training data composition, and evaluation tools.

• Alpamayo-R1 Model Release: NVIDIA has released the model weights and inference code for a pretrained 10B parameter model aimed at autonomous driving research, supporting the PhysicalAI-AV dataset and AlpaSim simulation.

• Gemini 3 Pro Model Integration: Google has added the model to the Venice API, enabling high-precision multimodal inference for text, image, and code tasks.

• Grok 4.1 Fast Model Integration: xAI has added its best agent-style tool-calling model to the Venice API, suitable for practical scenarios like customer support and image analysis.

• Kimi K2 Thinking Model Integration: Moonshot AI has integrated its most advanced open-source inference model into the Venice API, supporting agent-style long-horizon reasoning.

• Codex CLI Feature Update: OpenAI has updated the CLI to v0.65, adding support for resuming, medium inference effort levels, and stable rendering, ideal for reviewing large modules or refactoring in terminal workflows.

• Understand AI Module Feature Update: SciTools has rebuilt the module to support larger downloaded models, new hosting options, and a built-in AI chat with code-aware context.