Jamba Mini
Language model · Jan 8, 2026
Jamba Mini 1.6 brings AI21's hybrid state-space and Transformer design down to 52 billion total parameters and 12 billion active. It retains a 256K context and grounded generation.
Vendor · Language models
Jamba Models: All 7 Models from AI21 Compared
AI21 has 7 models in the directory: 7 language models.
Models73 of them since January 2026
Cheapest API$0.20Jamba2 Mini, input per 1M tokens
Most popular50Jamba Reasoning 3B, popularity index out of 100
Summary
In the directory: 7 language models. 7 of them with open weights that can run locally.
All 7 models from AI21
Every model there ever was. Models with a profile lead to every benchmark, price and source.
7 models
Status
| Jamba Mini | Jan 8, 2026 | Active | n/a | n/a | n/a | 256K | 9 |
| Jamba2 Mini | Jan 8, 2026 | Open source | n/a | $0.20 / $0.40 | n/a | 256K | 33 |
| Jamba2-3B | Jan 8, 2026 | Open source | n/a | n/a | n/a | n/a | 47 |
| Jamba Reasoning 3B | Oct 8, 2025 | Open source | n/a | n/a | n/a | n/a | 50 |
| Jamba Large | Jul 3, 2025 | Active | n/a | n/a | n/a | 256K | 16 |
| Jamba Large 1.7 | July 2025 | Active | n/a | $2 / $8 | n/a | 256K | 16 |
| Jamba Mini 1.7 | n/a | Limited access | n/a | n/a | n/a | 256K | 28 |
The current Jamba models explained
Jamba2 Mini
Language model · Jan 8, 2026
Jamba2-Mini uses mixture of experts within AI21’s hybrid SSM-Transformer architecture. It activates 12 of its 52 billion parameters per token. It launched alongside Jamba2-3B.
Jamba2-3B
Language model · Jan 8, 2026
Jamba2-3B combines state-space and Transformer layers in a dense 3B model. AI21 targets answers grounded in supplied documents, including manuals and company policies. The compact version works without a reasoning model’s additional thinking process.
Jamba Reasoning 3B
Language model · Oct 8, 2025
Jamba Reasoning 3B pairs reasoning with a memory-efficient SSM-Transformer architecture. AI21 reports a KV cache eight times smaller than that of a conventional Transformer. At a 32K-token context, the provider measures 40 output tokens per second on an M3 MacBook Pro.
Jamba Large
Language model · Jul 3, 2025
Jamba Large 1.6 combines state-space and Transformer layers in AI21's hybrid architecture. Its 398 billion total parameters, 94 billion active, and 256K context come with grounded generation.
Jamba Large 1.7
Language model · July 2025
Jamba Large 1.7 focuses on stronger grounding in supplied context, according to AI21. In the company’s FACTS evaluation, its score rises from 0.758 in version 1.6 to 0.832. The hybrid SSM-Transformer architecture handles 256K context tokens.
Where does AI21 stand?
Models among the best 20 in the overall index
Anthropic5 models, 4 of them among the best ten
OpenAI5 models, 2 of them among the best ten
Google4 models, 2 of them among the best ten
Meta3 models, 2 of them among the best ten
DeepSeek1 models, 0 of them among the best ten
Moonshot AI1 models, 0 of them among the best ten
xAI1 models, 0 of them among the best ten
AI210 models, 0 of them among the best ten
Vendors compared
| Vendor | Models | Model types | Best language model | Top 10 | Open weights | New in 2026 |
|---|---|---|---|---|---|---|
| Anthropic | 33 | 33 language models | Claude Opus 5.5 (Rank 1 to 3) | 4 | 0 | 13 |
| OpenAI | 63 | 46 language models, 4 image models, 2 video models, 9 audio models, 2 specialist models | GPT-6 Astra (Rank 4 to 8) | 2 | 4 | 27 |
| 66 | 43 language models, 4 image models, 6 video models, 11 audio models, 2 specialist models | Gemini 4 Carbon (Rank 1 to 5) | 2 | 19 | 26 | |
| Meta | 32 | 29 language models, 1 image model, 1 video model, 1 specialist model | Muse Wonder Full (Rank 6 to 9) | 2 | 26 | 7 |
| DeepSeek | 23 | 23 language models | DeepSeek V4.1 Flash 1017 (Rank 13 to 20) | 0 | 23 | 9 |
| Moonshot AI | 12 | 12 language models | Kimi K3.1 Pro (Rank 10 to 13) | 0 | 10 | 7 |
| xAI | 16 | 12 language models, 2 image models, 2 video models | Grok 4.7 (Rank 18 to 24) | 0 | 1 | 12 |
| AI21 | 7 | 7 language models | not ranked yet | 0 | 7 | 3 |
Strengths
Jamba2 Mini from $0.20 per 1M input tokens
7 models with open weights
Gaps
No image models, video models or audio models in the directory
No language model with a rank in the overall index yet
1 models only with limited access or as a preview
Every AI21 model over time
By release date. Grey: retired models.
20253 models
20263 models
No documented date1 model
Releases in the last twelve months
0Dec 25
3Jan 26
0Feb 26
0Mar 26
0Apr 26
0May 26
0Jun 26
0Jul 26
0Aug 26
0Sep 26
0Oct 26
0Nov 26
3 releases since January 2026, by release date.
API prices
Prices per 1M tokens for the current language models, according to AI21 or the provider linked in each model profile.
| Current model | Input | Output | Context |
|---|---|---|---|
| Jamba2 Mini | $0.20 | $0.40 | 256K |
| Jamba Large 1.7 | $2 | $8 | 256K |
Frequently asked questions about AI21
What Jamba models are there?
The directory lists 7 Jamba models from AI21: 7 language models. It counts every model AI21 ever released, including retired ones. 6 are currently available, the newest being Jamba Mini, Jamba2 Mini and Jamba2-3B.
What is the latest Jamba model?
Jamba Mini, released on Jan 8, 2026. The timeline below shows every release by date.
How much do Jamba models cost?
Through the API per 1M tokens (input / output): Jamba2 Mini $0.20 / $0.40. The prices of all 2 current language models are in the price table.
Can I run Jamba models locally?
Yes, 7 of the 7 models have open weights. Each model profile lists how much memory it needs.
How large is the context window of the Jamba models?
Up to 256,000 tokens (≈ 380 A4 pages), which 5 of the current models reach, including Jamba Mini. Every model's value is in the model list.
Sources and data date
Data as of Nov 1, 2026