Cerebras SUPERNOVA 2025: Andrew Feldman on AI Speed and Reasoning Models

Cerebras' Latest AI Innovations

• 🚀 Cerebras has launched the Qwen 32B model, which is the fastest in the industry, performing 20 to 30 times faster than GPU offerings.
• 🧠 Qwen 32B is a reasoning model that behaves like leading closed-source models, offering smart, open-source capabilities.
• 💡 The company is pushing the envelope on speed to meet market demand for faster AI processing.

The Critical Role of Speed in AI

• ⏱️ Reasoning models require more computational work and more tokens as they iterate through developing, reviewing, and improving answers.
• ⚡ Fast performance is crucial for agentic workflows, enabling complex tasks to be completed in seconds rather than minutes.
• 🎯 While time to first token is important for voice applications, output speed is generally the best measure for most other applications.

Benchmarking AI Performance

• 📊 Artificial Analysis is highlighted as a reliable benchmarking firm that pings APIs daily to report real-world performance results.
• 🔍 This approach provides a realistic measure of what customers experience, unlike historic benchmarks that could be tuned for months.
• ✅ The goal is to make benchmarking as realistic as possible, mirroring actual customer performance rather than special, optimized versions.

Enabling Diverse AI Applications

• 🏥 Large enterprises like GlaxoSmithKline are using Cerebras's technology for agentic flows in drug discovery, generating hypotheses, and linking to automated labs and robotics.
• 🌱 A new class of AI entrepreneurs and startups are leveraging Cerebras's horsepower to develop innovative use cases.
• 🤖 Computer vision is identified as a significant consumer of inference compute, requiring substantial processing power.

Strategic Partnerships and Market Expansion

• 🤝 Cerebras achieved a hyperscaler win with Meta, powering their Llama 4 API service for fast token delivery.
• 💼 IBM also partnered with Cerebras to provide fast inference to their large enterprise clients, expanding reach from developers to major financial institutions.
• 🌍 Partnerships with companies like Perplexity, Mistral, G42, and AlphaSense are delivering blazing fast models globally.

Empowering AI Developers

• 🛠️ Cerebras is offering one million free tokens per day to developers, with no waitlist, to encourage innovation.
• 💡 The message to developers is to use inference to solve specific, real-world problems, moving beyond "AI for AI's sake."
• 🚀 This initiative aims to foster an ecosystem where developers can create niche, domain-specific SaaS applications powered by Cerebras's fast engine.