Last verified
EU / ISRAEL LABJAMBASSM-TRANSFORMER256K CONTEXTOPEN WEIGHTS

Jamba API Pricing

Jamba is the model family from AI21 Labs, an Israeli AI lab founded in 2017. Jamba pairs a Mamba state-space architecture with Transformer layers for a 256K-token context at a lower cost than comparable long-context models, and ships as Jamba Large and Jamba Mini.

Priced models
2
2 GA
Founded
2017
AI21 Labs - Tel Aviv
Cheapest input
$0.2/M
Jamba Mini 1.7 input
Max context
256K
Jamba Large 1.7
Billing
token PAYG
public list
Architecture
SSM + Transformer
Mamba hybrid
§ 01 / LINEUP

The full roster.

Side-by-side →
§ 02 / SHELF

All side-by-side.

Methodology →
Model Input /M Output /M Cached Context Max output Vision Tools Tier
Jamba Large 1.7 FLAGSHIP $2.00 $8.00 256K Frontier
Jamba Mini 1.7 $0.2 $0.4 256K Active
§ 03 / PRICE CURVE

Pricing across the lineup.

How AI21 Labs priced 2 models · JUL 25.

oldest → newest →
$10$7.5$5$2.5 $0.40 $0.20 $8 $2 Jamba Mini 1.7 JUL 25 Jamba Large 1.7 JUL 25
Input · newest $2/M
Output · newest $8/M
Each point is a model at its listed $/M price.

The Jamba story so far.

Releases and pricing from AI21's Jamba line on AI21 Studio. Sourced from AI21 blog and press, the pricing page, and Wikipedia - verified at publication.

2025
Jamba 1.7 released - the current Large and Mini generation on AI21 Studio
RELEASE
2025
Jamba 1.6 shipped - AI21 held Large at $2/$8 and Mini at $0.20/$0.40 across the refresh
RELEASE
AUG - 2024
Jamba 1.5 Large and Mini released under an open model license - first production SSM-Transformer hybrid at 256K context
RELEASE
2023
AI21 raised a $208M Series C at a ~$1.4B valuation, backed by Google and Nvidia
CORPORATE
§ 04 / ACCESS

Where to get it.

Methodology →
§ 04 / BEST FOR

Which AI21 Labs for what.

More scenarios →
Long-document RAG on a budget?
Jamba Mini 1.7
Jamba Mini
Frontier long-context reasoning?
Jamba Large 1.7
Jamba Large
Self-hosting open weights?
Jamba 1.5 (open license)
AI21 on HF
Cheapest 256K-context input?
Jamba Mini at $0.20/M
Jamba Mini
§ 06 / BACKGROUND

The company behind it.

www.ai21.com →

AI21 Labs was founded in 2017 in Tel Aviv by Amnon Shashua, Ori Goshen and Yoav Shoham. It is one of the oldest independent LLM companies, predating the current generative-AI wave with its Jurassic language models and the Wordtune writing assistant.

AI21's current focus is the Jamba family, the first production-grade model to combine a Mamba state-space architecture with Transformer attention. That hybrid design gives Jamba a 256K-token context window at lower memory cost than a pure Transformer, which AI21 positions for long-document RAG and enterprise workloads.

Jamba is served per token on AI21 Studio and on Amazon Bedrock, with Jamba 1.5 weights released under an open license. Large lists at $2/$8 and Mini at $0.20/$0.40, held flat across the 1.5, 1.6 and 1.7 refreshes.

AI21 raised a $208M Series C in 2023 at roughly a $1.4B valuation, backed by Google, Nvidia and others. It competes with Mistral and Cohere as an independent, enterprise-focused alternative to the largest US labs.

§ 07 / COMPETITORS

Other frontier labs.

All providers →

Jamba pricing questions.

Practical questions about AI21 Jamba API pricing, with the published per-token rates kept separate from workload assumptions.

Q · 01 How much does the AI21 Jamba API cost? +
On AI21 Studio, Jamba Large is $2/M input and $8/M output, and Jamba Mini is $0.20/M input and $0.40/M output. Both support a 256K context. Verified on the AI21 pricing page 2026-07-11.
Q · 02 What is the difference between Jamba and a Transformer model? +
Jamba uses a hybrid SSM-Transformer (Mamba) architecture. The state-space layers make long contexts cheaper in memory, which is how AI21 offers a 256K-token window at a lower price than many pure-Transformer long-context models.
Q · 03 Has Jamba pricing changed across versions? +
No. AI21 has held Jamba Large at $2/$8 and Jamba Mini at $0.20/$0.40 across the 1.5, 1.6 and 1.7 releases, so upgrades did not change the per-token cost.
Q · 04 Can I self-host Jamba? +
Yes for the older generation - AI21 released Jamba 1.5 Large and Mini weights under the Jamba Open Model License. The current 1.7 models are served through the API rather than as open weights.
Q · 05 Is there a prompt-cache discount? +
AI21 does not publish a separate cache-read token price for Jamba, so this site keeps cached input equal to base input. Jamba is also available on Amazon Bedrock with AWS billing.