October 8, 2025
Introducing Jamba Reasoning 3B
Release Date: October 8, 2025What’s New:
We’re excited to announce Jamba Reasoning 3B, a compact open-source reasoning model that redefines what’s possible on-device.Built on AI21 Labs’ novel SSM-Transformer hybrid architecture, this 3B-parameter model delivers 2–5× efficiency gains over competitors while achieving leading intelligence benchmarks.Key Highlights
Smarter, faster, smaller: Hybrid SSM-Transformer efficiency enables long-context reasoning with minimal compute.Runs anywhere: Deploy locally on your phone or computer via llama.cpp or LM Studio.Enterprise-ready: Supports intelligent agentic workflows, reducing cloud costs and latency.Get Started:
- Check out the AI21-Jamba-Reasoning-3B and Jamba-Reasoning-3B-GGUF cards.
-
Read the full blog post.
September 18, 2025
You Can Now Use Requirements When Working With Tools!
Release Date: September 18, 2025What’s New:
AI21 Maestro now supports Requirements with both File Search and Web Search. This allows you to define explicit constraints, such as format, tone, or content rules. AI21 Maestro will refine outputs to meet them, even when external tools are used for additional context.What This Means for You:
- More control over how responses are structured and styled
- Greater consistency and reliability, even when using external tools
- Streamlined workflows with outputs that meet your standards the first time
September 18, 2025
Deprecation Notice: tool_resources Parameter
Release Date: September 18, 2025What’s New:
Thetool_resources parameter in POST /studio/v1/maestro/runs will be deprecated inOctober 2025 . Please use the tools parameter instead.For more information, check the
tools parameter documentation.September 16, 2025
New Data Connectors Now Available for Enterprise Customers!
Release Date: September 16, 2025What’s New:
- Available Data Connectors for Enterprise Users:
- Amazon S3
- Box Confluence
- Dropbox
- Google Drive
- Browse Website
- Intercom
- OneDrive
- Microsoft Teams
- Notion
- ServiceNow
- SharePoint
- Zendesk
August 7, 2025
Mistral models are now available in AI21 Maestro!
Release Date: August 7, 2025What’s New:
You can now access themistral-small, mistral-8x7b and mistral-7b models directly in Maestro’s API and playground. These models use AI21’s API key, so you don’t need to provide your own.See the full list of available models.
July 8, 2025
AI21 Maestro Is Now Live!
Release Date: July 8, 2025We’re excited to introduce AI21 Maestro.What’s New:
AI21 Maestro is an advanced AI system for rapidly building and deploying knowledge agents that automate complex, high-value business tasks. Built for enterprise use, it combines retrieval-augmented generation (RAG), real-time validation, and dynamic execution planning to deliver accurate, reliable outputs aligned with user-defined goals and constraints.Validated Output ensures that generated responses meet explicit user-defined requirements, such as formatting, content accuracy, tone, or domain-specific rules. Rather than relying solely on prompting, AI21 Maestro enforces these constraints by validating and automatically refining outputs using computational resources.AI21 Maestro RAG features a robust RAG engine that enhances answer reliability by grounding model outputs in enterprise knowledge sources. It supports both semantic file search and web search, enabling contextual, fact-based responses from internal documents or real-time external content.July 8, 2025
Google Drive Connector Now Available for Enterprise Customers!
Release Date: July 8, 2025You can now seamlessly import and sync your Google Drive files directly into your File Library, enabling faster, more integrated workflows.Key Benefits
-
Smart File Support – Automatically imports commonly used formats:
.pdf,.docx,.txt,.html, and.md. - Real-Time Sync – Edits, renames, and newly added files in Google Drive are instantly reflected in the platform.
- Setup Tip – First-time connections may take longer for large accounts, as content is scanned and filtered during setup. Our use of information received from Google Workspace APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. We do not use data obtained from these APIs to develop, improve, or train any machine learning or AI models beyond the user’s own personalized experience. Want to learn more? Talk to our sales team.
July 3, 2025
Introducing Jamba 1.7!
Release Date: July 3, 2025We’re excited to introduce the release of Jamba 1.7!What’s New:
- Smarter Answers with Enhanced Grounding: Jamba 1.7 now delivers more complete and accurate responses by better understanding context and focusing on what matters. Whether you’re tackling complex questions or seeking precise insights, Jamba 1.7 is an ideal fit for question-answering and instruction-following tasks.
- Faster and More Efficient: Thanks to optimized configurations and quantization, Jamba 1.7 is delivering top-tier performance without sacrificing quality.
- Self-Hosted Ready: Want more control? Jamba 1.7 is now available for self-hosted deployment, giving teams flexibility and scalability in their own environments.\
Get Started:
Access via AI21 Studio or Hugging Face - Jamba Large 1.7, Jamba Mini 1.7.
March 22, 2025
Introducing Jamba 1.6!
Release Date: March 6, 2025We’re pleased to announce the release of Jamba 1.6.Key Updates:
- Enhanced Model Quality: Jamba Large 1.6 outperforms leading open models from Cohere, Meta, and Mistral on quality (Arena Hard) and speed.
- Long context processing: With a 256K context window and hybrid SSM-Transformer architecture, Jamba excels on efficiently and accurately processing long contexts, outperforming leading open model competitors on RAG and long context QA benchmarks
- Secure Deployment: Available via AI21 Studio (SaaS) or to download from Hugging Face and deploy privately (VPC/on-prem) from Hugging Face. More deployment options coming soon.
- Improved Efficiency: Faster response times with high accuracy.
Get Started:
Access via AI21 Studio or Hugging Face - Jamba Large 1.6, Jamba Mini 1.6.