Easy AI Daily News Roundup | December 17, 2025

A curated digest of AI industry news for December 17, 2025, compiled from the Easy AI Daily report.

Model Releases and Updates

• Xiaomi MiMo-V2-Flash: 309B-parameter MoE model with 15B active parameters, 150 tokens/s inference, and 256K long context. Achieves state-of-the-art results on SWE-Bench and is available for free on OpenRouter. (Model details | OpenRouter)
• OpenAI GPT Image 1.5: New core model for ChatGPT Images with more precise image editing and faster generation; ranked #1 on multiple benchmarks and available to all ChatGPT users and via API. (Announcement | API docs)
• NVIDIA Nemotron-Cascade: 8B/14B-parameter open-source models using Cascade RL, achieving 43.1% pass@1 on SWE-Bench Verified. (Details)
• Meta SAM Audio: Open-source model that separates specific sounds from complex audio using text, visual, or time-span prompts; includes models, benchmarks, and a paper. (HF collection | Announcement)
• AllenAI Molmo 2: Video multimodal model built on SigLIP2 and Qwen3, leading open-source models on video pointing/counting tasks; Apache-2.0 licensed. (Announcement)
• Apple SHARP: Generates 3D Gaussians from a single image in 1 second — claimed ~1000x faster than diffusion models with better perceptual fidelity. (Paper summary)
• Claude Code 2.0.70: Anthropic ships 13 CLI changes, 3x memory usage improvement, and input-clearing fixes. (Changelog)
• Manus.im 1.6: Released to all users with platform improvements (details undisclosed). (Announcement)

Benchmarks and Performance

• MiMo-V2-Flash scores 73.4% on SWE-Bench Verified (71.7% on multilingual tasks), outperforming comparable models. (Performance details)
• GPT Image 1.5 ranks #1 on LMArena (1277), Design Arena (1344), and AA Arena (1272). (LMArena thread)
• Google's FACTS leaderboard launched: Gemini 3 Pro leads at 68.8%; multimodal tasks remain challenging (~47%). (Thread)
• OpenAI launches FrontierScience, an open benchmark for PhD-level scientific reasoning across physics, chemistry, and biology. (Announcement)

Open Source and Ecosystem

• Nemotron 3 Nano now available via Ollama, MLX, and LM Studio for local deployment. (Ollama)
• Mistral Small Creative, an experimental model, is live on OpenRouter at $0.10/$0.30 pricing. (Model page)
• Unsloth AI community testing finds Nemotron 3 Nano outperforms Qwen3 30B with lower failure rates and faster speeds. (HF model)
• Eleuther community members introduce Synthema, a meta-language for meaning compression into shorter symbolic syntax.

Multimodal and Audio

• MiniMax VTP: Open-source visual tokenizer improving diffusion model generation quality with no extra compute. (Announcement)
• Runway Gen-4.5 is now available to all paid users. (Announcement)

Tools and Infrastructure

• vLLM releases a Rust-based, KV-aware load balancer with consistent hashing, retries, and Kubernetes discovery. (Announcement)
• OpenHands ships a production-oriented agent SDK supporting tool calling and reasoning. (Announcement)
• Cline migrated to Vercel's AI Gateway, cutting error rates and improving P99 latency by 10–40%. (Announcement)
• tinygrad now immediately closes AI-generated PRs from unknown contributors, requiring contributors to understand every line of code. (Community notice)

Community and Discussion

• Terence Tao argues current AI is "artificial general intelligence" in name only, relying on stochastic or brute-force methods and falling short of human intelligence. (Source)
• The MI6 chief warned that tech giants' influence rivals governments, calling for urgent regulation to counter disinformation risks. (Source)
• A Reddit user reports quitting nicotine and gaming addiction within a week using ChatGPT conversations. (Reddit post)
• LMArena launched a YouTube channel and a December AI Generation Contest (theme: Holiday Celebration, submissions close December 30).

Discord Community Highlights

• BASI Jailbreaking: GPT-5 mini system messages, jailbreak prompts, and a fully jailbroken DeepSeek discussion.
• LMArena: GPT Image 1.5 version differences, Nano Banana Pro performance drops, Gemini 3 Flash latency, and model censorship debates.
• Unsloth AI: GRPO vs DPO, Nemotron vs Qwen comparisons, Colab H100 GPUs, and SAM Audio licensing questions.
• Cursor Community: HTTP 401 errors on suggestions, token billing discrepancies, and agent window defaults.
• OpenRouter: Announcements for free MiMo-V2-Flash, Mistral Small Creative, and Black Forest Lab FLUX.2 Max.

📌 Source: Easy AI Daily 🤖 Compiled by: AI assistant