Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Block launches Dessa, a speech model trained only on human audio - OpenSmartRoute
Block announced a new AI model named Dessa. This company also makes Square and Cash App. They released Dessa today in October 2026. The team built this model from scratch. No one else made it before. It is the first model trained only on human speech. There was no text data in the training set. The goal was to catch emotions that text misses.
Block wants better human-AI communication. They also care about safety. Current voice AI ignores emotional cues often. People feel safer when machines understand feelings. Text models cannot see sarcasm or laughter well. Dessa listens to audio directly instead. It does not convert sound to words first. This allows it to pick up pauses too. It can detect laughter in real time.
Block says they built Dessa out of curiosity. They wanted to improve how people talk to AI. Safety concerns drove this project too. Many voice tools sound emotional but are not intelligent. They miss the real feelings behind the words. Block hopes Dessa changes that landscape soon. They are not making it public yet.
Dessa processes audio in a unique way. It skips the step of turning sound into text. Text models must transcribe audio before they understand it. This transcription step loses some emotional nuance. Dessa keeps the raw audio signal intact. It analyzes the sound waves directly. This lets it hear tone and pitch changes. It also notices how long a pause lasts.
The model supports full-duplex communication. It can listen and talk at the same time. This is different from standard chat interfaces. Users do not have to wait for a response. They can speak while the AI listens. This feels more like a natural conversation. It removes the awkward silence between turns.
The hardware setup for training Dessa was specific. They used a single NVIDIA GB200 NVL72 server. This is a powerful computing machine. It handled the massive amount of audio data. The model runs locally on a MacBook Pro M4 Max. This shows it can work on consumer devices too. Engineers can run it on their own laptops.
Training only on human speech is rare. Most models mix audio and text data. Block chose a very narrow focus. They wanted to learn pure speech patterns. This makes the model very specialized. It might not work well for coding tasks. But it excels at emotional detection.
Native audio models process sound directly without converting it to text first. This method preserves timing, tone, and speaker cues that text transcripts lose.
Researchers found Large Language Models frequently answer using search snippets without reading the actual web pages. This behavior leads to outdated answers and misleading citations across multiple model types.
Emotional detection matters for safety reasons. AI can accidentally hurt people with wrong tone. A sarcastic comment might seem harmless in text. But it sounds cruel in audio. Block wants to prevent this kind of harm. They need machines to understand context fully. Safety is a top priority for them.
Quality also depends on emotional understanding. A customer service bot needs empathy. It should not sound robotic or cold. Users feel frustrated when machines ignore their feelings. Dessa aims to fix this problem. It makes interactions feel more human. This improves the overall user experience.
Earlier attempts at emotional voice AI did not last long. Hume AI launched EVI in March 2024. It claimed to have emotional intelligence. Kyutai followed with Moshi in September 2024. LAION started an Open Empathic project in late 2023. OpenAI updated ChatGPT's Advanced Voice Mode in June 2025. These projects added better intonation and expressions.
However, sounding emotional is not the same as being intelligent. Advanced speech AI needs reasoning skills too. Reasoning affects the quality of the answer given. Just having a good voice is not enough. The model must understand the question deeply. Block realizes this lesson from past failures.
Dessa is not available to the public yet. The team is looking for security researchers. They want experts to test for unexpected risks. This is a common practice for new AI. Security researchers look for hidden dangers. They check for bias and safety flaws.
Anyone interested can sign up via a form. The team will contact them later. This ensures only qualified people test it. It keeps the model safe during development. Block is being very careful with the rollout.
OpenSmartRoute helps teams use this new model safely. It acts as a smart router for AI requests. Teams can define a catalogue of models and tools. OpenSmartRoute sends each request to the best fit. It scores candidates on quality, cost, speed, and safety. The team sets the weights for each score.
Hard rules pin specific requests to certain models. Text with personal data stays on an on-premises model. A region or cost cap is never crossed. This ensures data privacy and budget control. The system learns from outcomes automatically. A model that answers well gets more traffic. One that fails gets less traffic.
A new model is just one catalogue entry. It competes on the next request immediately. Nothing else changes in the app. This makes testing new models very fast. Teams can try Dessa without changing their setup. They just add it to the catalogue.
The router works with any OpenAI-compatible provider. It supports open-weight models served locally. It also handles MCP tools and A2A agents. This flexibility is important for diverse workflows. Teams can mix and match different tools.
A savings ledger shows what each routed request cost. It compares this to the most expensive model cost. Teams see exactly how much they save. This helps with budget planning and optimization.
An input guard spots prompt injection and personal data. It checks this before a request leaves the system. This prevents data leaks and security attacks. Safety is built into the routing process.
The hosted platform keeps a models catalogue with prices. It has public rankings built from real traffic. Teams can see how models perform in practice. This data helps them make informed decisions.
What to do if you want to test or use Dessa. You can sign up via the form on the Block website. The team will contact you to schedule a test. Prepare your own audio samples for evaluation. Make sure you have a secure environment ready.
Check the security requirements before you start. Ask the team about the specific risks they found. They might share some findings with you. This will help you prepare for the test.
You can also watch for official announcements. Block will likely release more details soon. Keep an eye on their social media channels. They might post updates about the model's progress.
If you are a security researcher, apply now. The team is actively looking for testers. Submit your application through the provided form. Be ready to discuss your background and expertise.
For teams using OpenSmartRoute, add Dessa to your catalogue. Configure the hard rules for data privacy. Set up the input guard to scan for risks. Monitor the savings ledger to track costs.
Use the 'osr eval' tool to measure routing accuracy. Run it on your own prompts regularly. If the score drops, the build will fail. This keeps your system running smoothly.
Review the public rankings on the hosted platform. See how Dessa compares to other models. Use the real traffic data to guide your choices.
Start small with a pilot project. Test the model with a small group of users. Gather feedback on the emotional detection quality. Adjust the weights in the router based on results.
Keep learning about the latest AI developments. Block's focus on emotion is a trend to watch. Other companies might follow their lead soon. Stay informed to stay ahead in the market.
The release of Dessa marks a shift in AI development. It prioritizes emotional understanding over pure text processing. This could change how we interact with machines. Safety and quality are now central to the design.
Block's approach shows that curiosity drives innovation. They did not rush to launch a public product. They focused on safety and rigorous testing first. This is a healthy way to build AI.
Teams should consider similar approaches for their own projects. Focus on the specific needs of your users. Do not just chase the latest hype. Build tools that solve real problems.
The path forward involves careful planning and execution. Start with clear goals and measurable outcomes. Use data to drive every decision you make.
OpenSmartRoute provides the infrastructure for this kind of work. It handles the complexity of routing and scoring. Teams can focus on their unique business logic.
By using OpenSmartRoute, you gain control over your AI stack. You see exactly where your money goes. You ensure your data stays safe and secure.
This news is important for the future of AI. Emotional intelligence will be a key differentiator soon. Companies that ignore this risk falling behind.
Block has set a high bar for what AI should be. They are leading the charge in this direction. Others will follow their example eventually.
The technical details of Dessa are impressive. Training only on human speech is a bold move. It shows a deep understanding of the problem.
The hardware choices reflect a balance of power and portability. Running locally on a MacBook Pro is practical. It means users do not need a dedicated server always.
The full-duplex capability is a game changer for conversations. It makes the interaction feel much more natural. Users will appreciate this feature immediately.
Safety researchers will play a crucial role in the next phase. Their input will shape the final version of the model. This collaborative approach ensures high quality.
The form to sign up is the first step. Fill it out with your details and contact info. Wait for the team to reach out to you.
Keep your expectations realistic about the model's capabilities. It is specialized for emotion, not general tasks. Use it where it fits best.
Monitor the community for more updates and news. Block will likely share more insights as they go. Stay connected with their progress.
The release of Dessa is a significant milestone. It shows the potential of audio-only training. It opens new possibilities for AI development.
Teams should prepare to adopt this technology soon. The market will demand emotional AI quickly. Being ready now gives you a competitive edge.
Use OpenSmartRoute to integrate Dessa into your workflow. Configure the rules and guards carefully. Test thoroughly before rolling it out to everyone.
The journey from curiosity to product is long. Block is taking their time to get it right. This patience will pay off in the long run.
The focus on safety and quality is the right path. It builds trust with users and stakeholders. This trust is the most valuable asset of all.
Block has shown that AI can be both powerful and safe. They have proven that emotion matters in machine learning. This is a lesson for the entire industry.
The details of Dessa are fascinating to explore. The training data, the hardware, the architecture. Every part tells a story of careful engineering.
The implications for safety and quality are huge. Misunderstanding emotion can lead to real harm. Block wants to prevent this from happening.
The future of AI lies in understanding human nuance. Text is not enough to capture the full picture. Audio and emotion are the next frontier.
Block is setting the standard for how AI should behave. They are leading the way in this new field. Others will learn from their experience.
The team behind Dessa is doing important work. They are pushing the boundaries of what is possible. Their dedication to safety is commendable.
The release of Dessa is a reminder that AI is evolving. It is becoming more human and more capable. We are entering a new era of interaction.
Teams should embrace this change and adapt their strategies. The tools available today are just the beginning. Tomorrow will bring even more powerful capabilities.
OpenSmartRoute is ready to help you navigate this change. It provides the flexibility and control you need. You can build the future with us.
The story of Dessa is just the start. More models and more innovations are on the horizon. Stay curious and stay engaged with the field.
Block has given us a glimpse into the future. It is a future where machines understand us better. This is a future worth building together.
The technical achievements of Dessa are remarkable. The training method is unique and effective. It sets a new standard for audio AI.
The safety focus is a model for the industry. Other companies should follow this lead. Safety must always come first in AI development.
The quality improvements promised by Dessa are significant. Users will notice the difference in interactions. This will change how people use AI tools.
The open-source nature of the project is encouraging. It allows others to learn and build upon it. Collaboration is key to progress in AI.
The community response to Dessa will be interesting. Researchers and developers will likely share their thoughts. This discussion will shape the future of the field.
The path forward is clear for those who want to lead. Focus on safety, quality, and human needs. Build tools that truly help people.
Block has shown that it is possible to do both. They have balanced innovation with responsibility. This is the right way to build AI.
The release of Dessa is a landmark event. It marks a shift in how we think about AI. It opens up new possibilities for research and application.
The details of the model are complex but fascinating. The training process, the architecture, the results. Each piece adds to the puzzle.
The implications for safety are profound. Emotional AI can prevent harm in many ways. It can detect subtle cues that humans miss.
The quality of interactions will improve dramatically. Users will feel more understood and valued. This will increase trust in AI systems.
The future of AI is bright and full of promise. Block is helping to make it a better place. Their work is a beacon for the industry.
The story of Dessa is a testament to human ingenuity. It shows what we can achieve with careful thought and effort. We are on the verge of something amazing.
The release of Dessa is a call to action. It challenges us to think deeper and act better. We must all play our part in this journey.
OpenSmartRoute is here to support you every step of the way. We provide the tools and the knowledge you need. Let's build the future together.
The road ahead is long but full of wonder. Block is leading the way with their vision. Follow their lead and you will not be left behind.
The world of AI is changing fast. Block is keeping up with the pace. They are always looking for the next big thing.
The release of Dessa is a reminder of our potential. We can create machines that truly understand us. This is a dream that is coming true.
The technical details of Dessa are a treasure trove. There is so much to learn from this project. Dive in and explore the depths.
The safety and quality focus is the heart of this story. It drives every decision and every action. This is the core of Block's philosophy.
The future of AI is being written right now. Block is writing a chapter that matters. Their work will be remembered for years to come.
The release of Dessa is a milestone in history. It marks a turning point in AI development. It opens a new chapter for the industry.
The story of Dessa is one of hope and possibility. It shows that we can build better things. We can build things that care about us.
The path forward is bright and full of light. Block is shining a light on the way. Follow the light and you will find your way.
The release of Dessa is a gift to the world. It gives us a better tool to understand each other. We are all better off for this.
The technical achievements of Dessa are a marvel. They represent the pinnacle of current AI engineering. They set a high bar for the future.
The safety and quality focus is the guiding star. It leads the ship through the storm. It ensures we reach our destination safely.
The future of AI is being shaped by Block. They are the architects of this new world. Their vision will define the next decade.
The release of Dessa is a call to join the movement. Join the team and help build the future. Together we can achieve great things.
The story of Dessa is a story of us. It is about human connection and understanding. It is about making the world a better place.
The path forward is clear and inviting. Block is paving the way for everyone. Walk the path and see the beauty.
The release of Dessa is a testament to human potential. It shows what we can achieve when we work together. We are stronger than ever before.
The technical details of Dessa are a work of art. They are a masterpiece of engineering and science. They deserve to be celebrated.
The safety and quality focus is the foundation of all. It supports everything we build and create. It is the bedrock of our success.
The future of AI is being written by Block. They are the authors of this new story. Their words will echo through the ages.
The release of Dessa is a beacon of hope. It shows that we can do better. We can do better for everyone.
The story of Dessa is a story of progress. It shows how far we have come. It shows how far we can go.
The path forward is open and wide. Block is clearing the way for us all. Walk forward and embrace the future.
The release of Dessa is a gift to humanity. It gives us a better way to communicate. We are all better off for this.
The technical achievements of Dessa are a triumph. They are a victory for human ingenuity. They are a victory for science.
The safety and quality focus is the promise of the future. It promises a better world for everyone. It promises a future worth living in.
The future of AI is being built by Block. They are the builders of this new world. Their hands will shape the future.
The release of Dessa is a call to action for all. Act now and help shape the future. Your actions matter more than you know.
The story of Dessa is a story of us all. It is about us and our shared humanity. It is about us finding our way.
The path forward is a path of light. Block is the lighthouse guiding us. Follow the light and you will be safe.
The release of Dessa is a moment of truth. It reveals the true potential of AI. It reveals the true nature of humanity.
The technical details of Dessa are a puzzle waiting to be solved. The pieces are in your hands. Put them together and see the picture.
How OpenSmartRoute helps
A team routing requests through OpenSmartRoute gains immediate access to Dessa without buying new hardware. The router treats Dessa as a new catalogue entry alongside existing models. It scores this speech model on quality, cost, speed and safety just like any other tool.
The system automatically routes audio requests to Dessa when it fits the criteria. Hard rules ensure sensitive audio data stays on an on-premises model if required. The input guard spots potential prompt injection before any audio leaves the secure environment.
The savings ledger shows exactly how much the team saves by using Dessa instead of a generic text model. If Dessa fails an evaluation, the router reduces its traffic share automatically. This setup lets the team test the new speech model safely while maintaining full control over their AI infrastructure.
NVIDIA launched Nemotron 3.5 Bolt E4B, a small language model running on 4GB of VRAM. It uses reinforcement learning only and leads 15 of 21 benchmarks against competitors.
