/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Tokyo-based Sakana AI, founded by two Google researchers, releases three Japanese language models built using “model merging”, which combines existing AI models

Sakana AI, a Tokyo-based artificial intelligence startup founded by two prominent former Google (GOOGL.O) researchers …

Reuters Anna Tong

Context & Ripple Effects

Sakana AI was launched in Tokyo by former Google researchers and had already positioned itself around smaller models through a $30M seed round focused on smaller AI models. These releases put that positioning into a concrete Japanese-language product and technical approach.

The models also precede later evidence that the company’s Japanese-model work attracted major backing, including its $100M Series A after unveiling models for Japanese speakers.

First-order effects

  • Japanese-language developers and prospective customers gain three new Sakana AI model options built by combining existing models rather than training each one from scratch.
  • Sakana AI gets an early product test of model merging as a differentiator, alongside its smaller-model focus.

Second-order effects

  • Japanese-focused model providers may need to compete not only on proprietary training runs but also on how effectively they adapt, combine, and deploy available models for local-language use cases.
  • For customers, model evaluation can shift toward task performance and integration fit, rather than treating a single standalone model as the only unit of purchase or deployment.

Third-order effects

  • If model merging proves repeatable, AI development could place more value on orchestration and specialization of existing models, widening the path for regional entrants without requiring a wholly new foundation model.
  • That would reinforce a two-track market: globally developed base models paired with locally tuned or assembled systems for language- and market-specific deployment.

The trend: This is one data point in the shift from monolithic foundation-model building toward hybrid systems that adapt and combine models for specific markets and tasks.

Discussion

  • @sakanaailabs @sakanaailabs on x
    Training foundation models require enormous resources. We can overcome this by working with the vast collective intelligence of existing models. @HuggingFace has over 500k models in dozens of modalities that, in principle, can be combined to form new models with new capabilities!…
  • @vkhosla Vinod Khosla on x
    Innovation in AI keeps going. Excited for this evolution from @SakanaAILabs. Their new approach of *evolving* new foundational AI models, which leverages on the vast ocean of open-source AI models out there, enables them to efficiently, and with little effort, produce powerful...
  • @sakanaailabs @sakanaailabs on x
    @huggingface As a 🇯🇵 AI lab, we wanted to apply our method to produce foundation models for Japan. We were able to quickly evolve 3 best-in-class models with language, vision and image generation capabilities, tailored for Japan and its culture. Read more in our paper https://arx…
  • @sakanaailabs @sakanaailabs on x
    Introducing Evolutionary Model Merge: A new approach bringing us closer to automating foundation model development. We use evolution to find great ways of combining open-source models, building new powerful foundation models with user-specified abilities! https://sakana.ai/... [v…
  • @graceisford Grace Isford on x
    Thrilled to announce @SakanaAILabs's first launch (< 2 months after seed funding!) of 3 SOTA evolutionary AI foundation models for the Japanese market @SakanaAILabs challenges the current paradigm of costly model development w/ resource-efficient, evolved models THREAD👇🇯🇵