Product updates, new features, and what we're building next.
Today we're launching the complete 325 API platform with 25+ pages: API Playground, Model Catalog, Activity Analytics, Key Management, Team workspaces, Comparison tool, SDKs page, Help Center, and more. The platform now rivals OpenRouter and exceeds Claude/ChatGPT in features. All built by a solo developer in under a week.
Our fastest tier uses Cerebras wafer-scale hardware (0.3s latency) with automatic Groq fallback. Perfect for HTML/CSS generation, SQL queries, shell scripts, and simple tasks. 100% accuracy on JavaScript, HTML, SQL, and Shell benchmarks.
Need real-time data? The Research API combines Tavily web search with DeepSeek V4 Pro synthesis. 5+ sources per query with inline citations. Perfect for fact-checking, competitive research, and deep dives.
Fixed an issue where Cerebras models would return reasoning-only output when max_tokens was set too low. Enforced a 500-token minimum for 325-fast tier to ensure complete responses with actual content.
Our OpenAI-compatible endpoint went live with 4 tiers: 325-fast, 325-balanced, 325-ultra, and 325-auto. 10+ frontier models behind one API key. Zero token markup. Built by a developer tired of managing 12 different API keys.