AI21 Labs
Jamba models, an agent-cost gateway and post-training from Tel Aviv
Overview
AI21 Labs is the Tel Aviv lab behind the Jamba long-context models (Jamba2 open under Apache 2.0) that restructured in 2026 around an Intelligent Gateway for cutting AI-agent token costs and post-training of small open models. Jamba API prices are still listed, but self-serve sign-up currently redirects.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Jamba model access
- Jamba Mini: $0.20 in / $0.40 out per 1M
- Jamba Large: $2 in / $8 out per 1M
- Unlimited seats
- Volume discounts and premium rate limits
- Private cloud hosting
- Dedicated account manager and AI consultancy
- Works with Claude Code, Codex and LangChain
What you can produce with AI21 Labs
- Intelligent Gateway for cutting AI-agent token costs (beta)
- Post-training of small open models for customer workloads
- 256K-context Jamba hybrid Mamba-transformer models
- Jamba2 Mini and 3B open weights under Apache 2.0
- Token-metered Jamba API (listed; sign-up currently redirects)
- ai21-python and ai21-typescript SDKs (last released 2025)
AI21 Labs is in transition. Once a general model provider with Jamba's long context and the Maestro agent system, it shed most of its staff in May 2026 after acquisition talks failed and now leads with two narrower services: a gateway, free in beta, that trims token bills for coding agents, and post-training that tunes small open models to customer workloads. Jamba2's open weights under Apache 2.0 are a genuine asset for teams that want to self-host long-context models. As a hosted API, though, it is hard to recommend for new projects: prices are still published, but self-serve sign-up redirects away and the SDKs have not been updated since 2025. Worth a look for enterprises chasing agent cost savings or custom small models; developers wanting a stable model API should choose a provider whose roadmap is clearer.
What we found
AI21 Labs, founded in Tel Aviv in 2017 by Amnon Shashua, Yoav Shoham and Ori Goshen, built its reputation on language models — the Jurassic series, then Jamba, a hybrid of Mamba and transformer layers with a 256,000-token context window — and on Wordtune, its writing assistant. It raised $155 million in a Series C in 2023, from investors including Google, Nvidia, Samsung Next and Thailand's SCB10X, at a $1.4 billion valuation, for $283 million in total.
2026 changed the company. According to Calcalist, acquisition talks with Nvidia and then Nebius ended without a deal, and in May AI21 cut its workforce from about 180 to about 70, signed a commercial partnership with Nebius and said it was discontinuing the sale of standalone models. Its website now leads with two services. The Intelligent Gateway, in beta and free during the preview, sits between AI agents such as Claude Code or Codex and their models to cut token spending — AI21 claims savings of up to 40%, and 68% in one code-review case. Post-Training tunes small open models to match frontier models on a customer's own workloads. Maestro, the agent-planning system it launched in 2025 and called the world's first, no longer appears in the site's navigation.
The models remain partly available. Jamba2, released in January 2026, comes in Mini and 3B sizes under the Apache 2.0 licence, and the pricing page still lists API rates — Jamba Mini at $0.20 per million input tokens and $0.40 output, Jamba Large at $2 and $8 — with a $10 seven-day trial. But the self-serve sign-up link currently redirects to the homepage, so new developers may need to contact sales; the official SDKs were last released in 2025.
AI21 counts Nebius, Wix and Fnac Darty among its partners and customers. Treat it today as a specialist in agent cost and model tuning with open-weight Jamba models, rather than a general model API to build on.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including AI21 Labs's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with AI21 Labs unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to AI21 Labs directly →
Spotted something out of date? Suggest an update →
More in LLMs & Chat
