Taalas
Transform AI models into efficient, silicon-embedded solutions.. [Contact for Pricing]
Last verified:
What is Taalas?
Taalas is a platform that transforms any AI deep learning model into custom silicon, creating what they call
Taalas pricing
Pricing model: Freemium
Taalas operates on a paid-only pricing model with no free plan and no free trial available. The company does not publicly list specific pricing tiers or costs on their website. API access is currently closed due to overwhelming demand. Interested customers must contact [email protected] or request API access through their form. The company has invested $30M in development and targets enterprise customers in the AI and semiconductor sectors.
Taalas pros
- 1000x more efficient than software AI models
- 17,000 tokens per second per user inference speed
- Converts any AI model to custom silicon
- Supports fine-tuning of hardcore models
- Apps written in human languages
- TSMC 6nm manufacturing process
- 20x cheaper to build than alternatives
- 16,000-17,000 tokens/sec per user on Llama 3.1 8B
- No HBM required for weights
- Complete model-to-chip conversion in 2 months
- Hard-wired silicon implementation
- Optimal hardware-native execution
- 10x faster than current state of the art
- 53B transistors on 815mm² chip
- 2.5 kW server power consumption
- Instantaneous inference capability
- Led by Tenstorrent co-founder
Taalas cons
- No free plan available
- No free trial offered
- API access currently closed to new requests
- Only paid pricing model
- Limited to specific AI models initially
- Custom silicon requires manufacturing lead time
- 24-person company with limited scale
- Private company with no public stock
- Data transfer to United States required
- Specialized hardware limits flexibility
Frequently asked questions about Taalas
What is Taalas?
Taalas is creating a platform for quickly turning any AI model into custom silicon. The resulting Hardcore Models are 1000x more efficient than their software counterparts running on traditional computers.
What is the Taalas HC1 chip?
The HC1 is a Technology Demonstrator that runs the Llama 3.1 8B model. It uses TSMC 6nm process, has 815mm² size with 53B transistors, operates on a 2.5 kW server, and delivers 17,000 tokens per second per user.
What is Taalas Foundry?
Taalas Foundry is the platform that automates the workflow to map deep learning models into optimized, hardware-native implementations. Any AI model can be made Hardcore through Taalas Foundry, and it can convert a new model into hardware on a custom chip in two months.
Do Hardcore Models support fine-tuning?
Yes, Hardcore Models created through Taalas support fine-tuning, allowing users to adapt the hardware-embedded AI models to their specific use cases.
How are apps written for Hardcore Models?
Apps for Hardcore Models are written in human languages rather than traditional programming code, making them more accessible to developers.
What is the performance of Taalas compared to competitors?
HC1 delivers 17k tokens per second per user on Llama 3.1 8B, achieving nearly 10x faster performance than current state of the art while costing 20x less to build. The HC1 is 1000x more efficient than software counterparts on GPUs like H200 and B200.
Is there a free tier or free trial?
No, Taalas does not offer a free plan or free trial. The pricing model is paid-only. However, there is a free chatbot interface available to try on their website.
How do I get API access?
API access requests are currently closed due to overwhelming demand. You can contact [email protected] to get in touch, but the API Access Request Form indicates they are no longer taking applications at this time.
Where is Taalas headquartered?
Taalas is headquartered in Toronto, Ontario, Canada. The company is privately held with 2-10 employees and operates in the semiconductor manufacturing industry.
Who founded Taalas?
Taalas is led by Ljubisa Bajic, who is a co-founder of Tenstorrent. The company has 24 dedicated people working on the platform and has spent $30M on development.