🤖 China 40B model breaks SWE-bench

Happy New Year! A Chinese quant firm just shipped a coding model that punches way above its weight, and the numbers are hard to ignore. Plus, I’ll show you how to edit images, generate videos, and create content using one tool.

In today’s AI newsletter:

• IQuest-Coder Hits SWE-Bench With Just 40B Params

• Instagram Quietly Accepts an AI-First Internet

• SoftBank Finishes Its $40B Bet on OpenAI

• How to Edit Images, Generate Videos, and Create Content Using One Tool

• 4 new AI tools worth trying

AI MODELS

🧠 IQuest-Coder Hits SWE-Bench With Just 40B Params

IQuest-Coder-V1, a new family of code LLMs, scored 81.4% on SWE-Bench Verified with ~40B parameters, putting it in territory usually reserved for much larger models.

• 81.4% on SWE-Bench Verified, plus strong results on BigCodeBench and LiveCodeBench

• Trained with a “code-flow” paradigm that learns from repo evolution, commits, and real dev workflows

• Two variants: Thinking (RL + deep reasoning) and Instruct (general coding & agent use)

• Native 128K context and a recurrent “Loop” variant to reduce deployment cost

SWE-Bench has become the gold standard for autonomous software agents. If a 40B model can reliably hit this level, it signals a shift: smaller, workflow-aware models may outperform brute-force scale. The coding-agent race just got more competitive, and more efficient.

AI CONTENT

📸 Instagram Quietly Accepts an AI-First Internet

Instagram head Adam Mosseri says AI imagery will soon be so ubiquitous that it’ll be easier to cryptographically verify real photos than detect fake ones, a major philosophical shift for social media.

• Mosseri argues AI-generated media will overwhelm detection systems as models get better at realism

• Proposes “fingerprinting real media,” with cameras cryptographically signing images at capture

• Admits AI labeling and watermarking are unreliable and getting worse over time

• Signals creators may need to post more “unflattering” or raw content to signal authenticity

This is a quiet surrender. Instead of policing AI slop, platforms like Instagram may offload truth verification to hardware makers and creators themselves. If “real” has to prove itself, the burden of authenticity shifts, and the internet’s default may soon be synthetic by design.

AI PARTNERSHIPS

💰 SoftBank Finishes Its $40B Bet on OpenAI

SoftBank has completed its entire ~$40B investment into OpenAI, finalizing one of the largest private tech bets ever and cementing OpenAI as the center of the AI infrastructure race.

• SoftBank sent the final ~$22B last week, taking its total investment to ~$41B and ~11% ownership

• Deal values OpenAI at a reported $260B pre-money

• Part of the capital supports Stargate, OpenAI’s massive AI infrastructure push with Oracle

• SoftBank sold its entire $5.8B Nvidia stake and acquired DigitalBridge for $4B to fund AI infra

• OpenAI has lined up $1.4T+ in long-term infrastructure commitments across chips and compute

The AI race is shifting from who builds the best model to who can finance, deploy, and scale infrastructure fastest. In the long run, compute availability may matter more than capability breakthroughs.

HOW TO AI

🗂️ How to Edit Images, Generate Videos, and Create Content Using One Tool

In this tutorial, I’ll show you a free all-in-one AI platform that brings multiple creative tools together in one place. From image generation and background removal to video tools and utilities.

🧰 Who is This For

• Creators who want quick AI outputs without juggling multiple websites

• Designers looking for fast assets and visual experiments

• Beginners who want to explore AI tools without technical knowledge

• Anyone who wants free, practical AI tools in one dashboard

STEP 1: Access the Tool

Start by opening your browser and go to toolfk.com. As soon as the dashboard loads, you’ll notice a wide collection of AI utilities, ranging from video generators to background removers.

You can use each of these tool for completely free. Once you are ready just click on the tool you want to use.

For this tutorial, focus on the top row of tools and click on “AI Image Generator.” This opens the image generation workspace.

STEP 2: Configure Your Settings

Inside the generator, you’ll see a clean, dark-themed interface designed to keep things simple.

The model will already be set to ZImage, which is the default and recommended option. This ensures fast generation with strong visual quality.

Next, set your image resolution. The default size is 1024 × 1024, which works perfectly for profile images, posts, and thumbnails. You can adjust the width and height if you need a different format.

If you’re writing prompts in another language, use the “Translate to English” button. The model performs best when prompts are in English, and this saves you from rewriting anything manually.

STEP 3: Write or Enhance Your Prompt

Click into the main text box and describe the image you want to generate. Be clear and specific about the subject, style, lighting, and framing. For example, you might describe a realistic portrait, a cinematic scene, or a stylized character.

At the bottom of the screen, you’ll notice prompt enhancement suggestions such as “highly detailed,” “cinematic lighting,” or “8k resolution.” Instead of typing these yourself, you can simply click “Apply Suggestions.” The tool automatically upgrades your prompt to improve image quality.

STEP 4: Generate and Download Your Image

Once your prompt is ready, click the “Generate Image” button. The image is created very quickly, and the result appears in the center of the screen.

Take a moment to review the output, paying attention to details like lighting, textures, and overall realism. If you’re happy with the result, click “Download Image” at the bottom to save it directly to your device.

China's draft AI rules have measures to protect kids, like time limits and parental consent, ban gambling promotion content, and add suicide chats restrictions.

Moonshot AI, maker of Kimi models, raised a $500M Series C at a $4.3B valuation led by IDG Capital, with Alibaba and Tencent participating.

Elon Musk says Neuralink plan to start “high-volume production” of brain-computer interface devices and move to an almost automated surgical procedure in 2026.

China is requiring chipmakers to use at least 50% domestically made equipment for adding new capacity, in a rule that is not publicly documented.

⚙️ IQuest-Coder-V1: China's new 40B Model crack 80% on SWE-Bench

🧑‍💻 MiniMax 2.1: Strong model for coding and building apps

🗣️ Chatterbox Turbo: Fast, natural-sounding text-to-speech model from Resemble AI

🤖 Manus: a top-level general AI agent, recently bought by Meta

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas