Sapien Weekly Digest

Hey there, Sapiens! Welcome to December, only a very short amount of time left in the year but we’re not done building just yet as we’re continuing to build Proof of Quality into the universal settlement layer for quality.

What we’ve been up to

Bithumb:
We’re officially live on Bithumb! Bithumb has listed SAPIEN on the KRW market, with trading opening on December 5, 2025, at 17:00 KST.

Sapien Team Takeover:
Today, we hosted another AMA on our Discord, this time with our Director of Solutions Architecture, Lukas Grapentine! The AMA was centered on Sapien’s Proof of Quality system, staking model and the future roadmap. In practice, any industry relying on human judgment or model evaluation should be able to anchor its truth, trust, and verification standards to Proof of Quality over time. Read the full AMA here!

🚀 What’s New in the App?

As we’re going through a massive overhaul of the app, I cannot share the full details quite yet - however, the team is hard at work and we will be able to present our ideas in public in the near future. For now, we’re aiming to provide you with a continuous flow of tasks, as well as opening up Quality Assurance for high reputation users.

Current task Overview:

1. 👩‍🍳 Gastro Tag – Easy / Intermediate / Hard

2. 👩‍🍳 Gastro Tag QA – Easy / Intermediate / Hard

3. 🧩 Logic Path Vietnamese QA

4. 👋 Hand Object Annotation

5. 📦 Breakfast Image Bounding Boxes

6. 🚗 Vehicle Model Mapping

New Tasks:

Gastro Tag and Gastro Tag QA are live NOW!

We’re opening up the Quality Assurance of labelled data to the rest of the community! Now you, as a Sapien with a high Reputation score, can check the work of your fellow Sapiens and confirm the quality of their work. This is the first task of multiple along these lines that we’re aiming to share with you going forward! We will be releasing audio upload for emotion datasets next week, along with a multiple choice analysis!

Where are we going?

A new Sapien Experience! Proof of Quality will be introduced in stages, with new features and modules coming online over time. It’s  an evolving protocol that will increasingly shape all interactions in the ecosystem.

A more formal roadmap is planned, transparency is our biggest goal as features move from internal development to public release.  High-reputation community members are expected to play a bigger role in quality control as Proof of Quality matures, especially in validation and calibration workflows.

To that end, we’re aiming to move beyond traditional labeling and into more complex, higher-value categories. Proof of Quality is the enabler here, because it provides the trust infrastructure required for higher-stakes workflows.

A major upcoming feature is dispute transparency. Contributors will be able to dispute QA feedback and gain visibility into how their work was evaluated, targeting issues such as identical images being judged inconsistently at different times.

Christmas! Stay tuned for Christmas time on the Discord! We’ll make sure it’s a space for everybody to come in and enjoy, regardless of what you celebrate and whether or not you do!

Our Voices in the World

"With infinite access to the internet came infinite noise."

Check out our Co-Founder Rowan Stone speak about what comes after the Internet we know, on an X Space with our friends at the Meteorite Collective, where he spoke about just how important a system like Proof of Quality is for anybody building AI models.

Just earlier today, we were also featured on Episode 5 of Veera’s The Financial Loop, giving Rowan an opportunity to go into detail on how Sapien is using AI training models to reshape onchain earnings, and what this means for everyday users.

Additionally, our Co-Founder Trevor Koverko was quoted in The Defiant speaking about the long-term outlook for the market and what to expect.

What else happened in AI?

The World outside of Sapien:

“Code Red” for OpenAI:
OpenAI is entering a self-declared “code red” period, shifting much of the company’s focus back to the product that made it a household name. In an internal memo to employees, CEO Sam Altman described this as “a critical time for ChatGPT” and outlined an aggressive surge plan that prioritizes core model quality, personalization, and multimodal capabilities over new commercial experiments. Near-term updates will favor personalization, stability, and image quality enhancements. A soon-to-ship reasoning model, widely linked to the Shallotpeat codename, is designed to perform better on tasks that require multi step logic, planning, and complex problem solving. Beyond the near term release, the memo describes a larger model upgrade codenamed “Garlic,” with a target window in 2026. Garlic is presented as a successor level system that could align with a GPT 5.2 or GPT 5.5 style release. The defining goal is to both increase capability and address prior pre-training issues that limited earlier models, including improved handling of noisy or low quality data, better objective design for reasoning tasks, and more robust strategies for long context understanding.

Mistral 3: open-weight frontier models at Apache 2.0:
Mistral AI launched the Mistral 3 family, a suite of open-weight large language models expanding its catalogue of Apache-licensed open-weight models. The lineup includes three small dense models at 3B, 8B, and 14B parameters and one large mixture-of-experts model totaling 675B parameters, with 41B active at inference, all distributed under Apache 2.0 licensing. The Small 3 models are positioned for fine-tuning, constrained compute environments, and robotics, whereas the larger model extends to multimodal reasoning with 256K token context, reaching parity with the best instruction-tuned open-weight models on general prompts and entering the top tier of the LMArena OSS rankings. All models include multilingual capabilities designed for global enterprise workloads. Nvidia collaborated to optimize the family across its GPU stack. Microsoft added Mistral Large 3 to its Foundry program for production workloads. Mistral positions this as a foundational release for scalable, sovereign AI adoption.

Meta’s AI licensing deals with major news publishers:
Meta has finalized a series of paid data licensing agreements with major news publishers to feed its AI assistant with verified content. The deals cover outlets including USA Today, CNN, Fox News, People, The Daily Caller, the Washington Examiner, and France’s Le Monde. The deals allow Meta to ingest archival and current news content into its AI systems. Meta AI will use the material to generate real-time news answers for users across Facebook, Instagram, WhatsApp, and Messenger, complete with links that drive readers back to originating outlets.
For Meta, these partnerships help improve the accuracy and credibility of its assistant across platforms. For media companies, they offer a new income stream and guaranteed visibility inside an increasingly AI-centric discovery environment. Meta’s approach suggests that news distribution is migrating from feeds to chat interfaces, with licensed data forming the backbone of answer engines.

See you next week!