Skip to main content

AI21 LabsLanguage modelOpen source

Jamba2 Mini

Jamba2 Mini has no overall index value yet because there are too few independent benchmark results.

Intelligencenot ranked yet

n/aOverall indexhigher is better

Speednot measured

n/ano measurement yet

PriceRank 14 of 103

Input $0.20 · Output $0.40

$0.25per 1M tokenslower is better

ContextRank 85 of 112

256Ktokens in context≈ 380 A4 pages

PopularityRank 336 of 342 language models

14out of 100new calculation

based on two data groups, mostly search, range 4 to 24

Summary

It is cheap: $0.25 per 1M tokens (input and output blended); a typical model costs $1.50.

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.

AI21 specifies a 256K-token context window. The weights use Apache 2.0 and are offered through AI21 Studio and Hugging Face.

Prices

Choose a price view

1 offer, prices in US dollars per 1M tokens, blended from 3 parts input and 1 part output.

ProviderInputOutputBlended
AI21 Labsdirect$0.20$0.40$0.25

Direct price according to the developer, checked Oct 29, 2026. Open the LLM price index

More price tiers and rules
Source pricing conditions
Input $0.20, output $0.40 per 1M tokens (retrieved 2026-10-29)Source

Unchanged since tracking began on Nov 1, 2026.

InputOutput

$0$0.20$0.40$0.60Nov 1, 2026Nov 20, 2026
Daily observed price in the LLM price index in US dollars per 1M tokens. The index uses a market price from several sources, which can differ from the developer's direct price.

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 29 published benchmark results

Page 1 of 3

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

GitChameleon 2.0 Jamba 1.6 Mini k=1 success rate (Show measurement, test conditions, and source)
Source value
18
Metric
k=1 success rate
Unit
%
Benchmark version
GitChameleon 2.0, ACL 2026
Category
code-agent
Direction
Higher is better
Sample
328
Source type
independent-evaluator
Evaluator
GitChameleon 2.0, ACL 2026 paper
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. GitChameleon 2.0, exact publicly named model variant. The paper evaluates 328 manually authored version-conditioned Python API-change problems with hidden tests. Model serving revision and complete decoding parameters are not specified. Jamba 1.6 Mini; Table 12 compares k=1 and k=3 retrieval; precision is printed only for k=3 because the paper states it equals recall at k=1.
Context
The table prints ± values but does not identify their statistical meaning or confidence level; no CI field is set. The evaluated serving revision is not pinned.
Source GitChameleon 2.0, ACL 2026 paper
GitChameleon 2.0 Jamba 1.6 Mini k=1 API hit rate (Show measurement, test conditions, and source)
Source value
35.4
Metric
k=1 API hit rate
Unit
%
Benchmark version
GitChameleon 2.0, ACL 2026
Category
code-agent
Direction
Higher is better
Sample
328
Source type
independent-evaluator
Evaluator
GitChameleon 2.0, ACL 2026 paper
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. GitChameleon 2.0, exact publicly named model variant. The paper evaluates 328 manually authored version-conditioned Python API-change problems with hidden tests. Model serving revision and complete decoding parameters are not specified. Jamba 1.6 Mini; Table 12 compares k=1 and k=3 retrieval; precision is printed only for k=3 because the paper states it equals recall at k=1.
Context
The table prints ± values but does not identify their statistical meaning or confidence level; no CI field is set. The evaluated serving revision is not pinned.
Source GitChameleon 2.0, ACL 2026 paper
GitChameleon 2.0 Jamba 1.6 Mini k=3 success rate (Show measurement, test conditions, and source)
Source value
29.3
Metric
k=3 success rate
Unit
%
Benchmark version
GitChameleon 2.0, ACL 2026
Category
code-agent
Direction
Higher is better
Sample
328
Source type
independent-evaluator
Evaluator
GitChameleon 2.0, ACL 2026 paper
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. GitChameleon 2.0, exact publicly named model variant. The paper evaluates 328 manually authored version-conditioned Python API-change problems with hidden tests. Model serving revision and complete decoding parameters are not specified. Jamba 1.6 Mini; Table 12 compares k=1 and k=3 retrieval; precision is printed only for k=3 because the paper states it equals recall at k=1.
Context
The table prints ± values but does not identify their statistical meaning or confidence level; no CI field is set. The evaluated serving revision is not pinned.
Source GitChameleon 2.0, ACL 2026 paper
GitChameleon 2.0 Jamba 1.6 Mini k=3 API hit rate (Show measurement, test conditions, and source)
Source value
40.4
Metric
k=3 API hit rate
Unit
%
Benchmark version
GitChameleon 2.0, ACL 2026
Category
code-agent
Direction
Higher is better
Sample
328
Source type
independent-evaluator
Evaluator
GitChameleon 2.0, ACL 2026 paper
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. GitChameleon 2.0, exact publicly named model variant. The paper evaluates 328 manually authored version-conditioned Python API-change problems with hidden tests. Model serving revision and complete decoding parameters are not specified. Jamba 1.6 Mini; Table 12 compares k=1 and k=3 retrieval; precision is printed only for k=3 because the paper states it equals recall at k=1.
Context
The table prints ± values but does not identify their statistical meaning or confidence level; no CI field is set. The evaluated serving revision is not pinned.
Source GitChameleon 2.0, ACL 2026 paper
GitChameleon 2.0 Jamba 1.6 Mini k=3 precision (Show measurement, test conditions, and source)
Source value
41.6
Metric
k=3 precision
Unit
%
Benchmark version
GitChameleon 2.0, ACL 2026
Category
code-agent
Direction
Higher is better
Sample
328
Source type
independent-evaluator
Evaluator
GitChameleon 2.0, ACL 2026 paper
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. GitChameleon 2.0, exact publicly named model variant. The paper evaluates 328 manually authored version-conditioned Python API-change problems with hidden tests. Model serving revision and complete decoding parameters are not specified. Jamba 1.6 Mini; Table 12 compares k=1 and k=3 retrieval; precision is printed only for k=3 because the paper states it equals recall at k=1.
Context
The table prints ± values but does not identify their statistical meaning or confidence level; no CI field is set. The evaluated serving revision is not pinned.
Source GitChameleon 2.0, ACL 2026 paper
GitChameleon 2.0 Jamba 1.6 Mini k=3 recall (Show measurement, test conditions, and source)
Source value
50.1
Metric
k=3 recall
Unit
%
Benchmark version
GitChameleon 2.0, ACL 2026
Category
code-agent
Direction
Higher is better
Sample
328
Source type
independent-evaluator
Evaluator
GitChameleon 2.0, ACL 2026 paper
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. GitChameleon 2.0, exact publicly named model variant. The paper evaluates 328 manually authored version-conditioned Python API-change problems with hidden tests. Model serving revision and complete decoding parameters are not specified. Jamba 1.6 Mini; Table 12 compares k=1 and k=3 retrieval; precision is printed only for k=3 because the paper states it equals recall at k=1.
Context
The table prints ± values but does not identify their statistical meaning or confidence level; no CI field is set. The evaluated serving revision is not pinned.
Source GitChameleon 2.0, ACL 2026 paper
LiveCodeBench (Show measurement, test conditions, and source)
Source value
9.92
Metric
accuracy
Unit
%
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI LiveCodeBench leaderboard
Status
active
Retrieved at
2026-10-19
Methodology
Independent evaluation. Vals AI's public LiveCodeBench leaderboard, exact named row “Jamba 1.6 Mini”. The table labels the score Accuracy. The page is marked updated 9/1/2026, but does not give this row's evaluation date, model/API revision, sampling parameters, reasoning effort, sample count, or benchmark version.
Context
The page-level update date is not treated as the run date. No standard error, N, confidence interval, or LiveCodeBench version is inferred for this row.
Source Vals AI LiveCodeBench leaderboard
TaxEval v2 (Show measurement, test conditions, and source)
Source value
44.6
Metric
Accuracy
Unit
%
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI Jamba 1.6 Mini model page
Status
active
Retrieved at
2026-10-24
Methodology
Independent evaluation. Vals labels this its Accuracy Rankings page for Jamba 1.6 Mini. The page gives AI21 Labs as the default provider, temperature 0.4, default Top P/Top K, and maximum output of 4,096 tokens, while warning that some benchmarks may use different providers or parameters.
Context
The excerpt does not define the displayed ± values or the adjacent numerator/denominator-like pairs, so neither is treated as a confidence interval or evaluation N. Benchmark-specific versions, splits and settings are not stated in this excerpt.
Source Vals AI Jamba 1.6 Mini model page
GPQA Diamond (Show measurement, test conditions, and source)
Source value
32.07
Metric
Accuracy
Unit
%
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI Jamba 1.6 Mini model page
Status
active
Retrieved at
2026-10-24
Methodology
Independent evaluation. Vals labels this its Accuracy Rankings page for Jamba 1.6 Mini. The page gives AI21 Labs as the default provider, temperature 0.4, default Top P/Top K, and maximum output of 4,096 tokens, while warning that some benchmarks may use different providers or parameters.
Context
The excerpt does not define the displayed ± values or the adjacent numerator/denominator-like pairs, so neither is treated as a confidence interval or evaluation N. Benchmark-specific versions, splits and settings are not stated in this excerpt.
Source Vals AI Jamba 1.6 Mini model page
LegalBench (Show measurement, test conditions, and source)
Source value
69.73
Metric
Accuracy
Unit
%
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI Jamba 1.6 Mini model page
Status
active
Retrieved at
2026-10-24
Methodology
Independent evaluation. Vals labels this its Accuracy Rankings page for Jamba 1.6 Mini. The page gives AI21 Labs as the default provider, temperature 0.4, default Top P/Top K, and maximum output of 4,096 tokens, while warning that some benchmarks may use different providers or parameters.
Context
The excerpt does not define the displayed ± values or the adjacent numerator/denominator-like pairs, so neither is treated as a confidence interval or evaluation N. Benchmark-specific versions, splits and settings are not stated in this excerpt.
Source Vals AI Jamba 1.6 Mini model page
MMLU Pro (Show measurement, test conditions, and source)
Source value
30.28
Metric
Accuracy
Unit
%
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI Jamba 1.6 Mini model page
Status
active
Retrieved at
2026-10-24
Methodology
Independent evaluation. Vals labels this its Accuracy Rankings page for Jamba 1.6 Mini. The page gives AI21 Labs as the default provider, temperature 0.4, default Top P/Top K, and maximum output of 4,096 tokens, while warning that some benchmarks may use different providers or parameters.
Context
The excerpt does not define the displayed ± values or the adjacent numerator/denominator-like pairs, so neither is treated as a confidence interval or evaluation N. Benchmark-specific versions, splits and settings are not stated in this excerpt.
Source Vals AI Jamba 1.6 Mini model page
IFBench (Show measurement, test conditions, and source)
Source value
0.38
Metric
IFBench
Unit
ratio
Category
source-specific
Direction
Higher is better
Source type
vendor-reported
Evaluator
AI21 announcement/HF chart
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Same four normalized benchmark components and Enterprise Reliability composite; separate 100-prompt blind human comparison.
Context
Source-specific observation; it is not a shared comparison cohort. AI21 self-reported comparison, not an external evaluator record
Source AI21 announcement/HF chart

Page 1 of 3

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

The 100 most popular models of this category, 0 to 100, higher is more popular

20 of 101 models

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 20, 2026

Jamba2 MiniCode Llama 70B Instruct 14Seed 2.0 Lite 14DeepSeek V4 Pro 0423 14Reka Edge 14

Method changes in the timeline, a jump at these points is no market event:

  1. 1November 1, 2026: Value only with search or usage data, 30-day attention
  2. 2November 1, 2026: Method version 4 (common scale, shrinkage)
  3. 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

Cost for your load

1,000 requests with 2,000 input and 500 output tokens cost $0.6000 with Jamba2 Mini. Enter your own load in the cost calculator

All specifications

Architecture

Model type
Open source
Source: Model type
Active parameters
12B
Source: Active parameters
Parameters
52 billion total, 12 billion active
Source: Parameters

Input and output

Maximum output
4,096 tokens
Source: Maximum output
Context window
256K
Source: Context window
Input
Text (retrieved 2026-10-29)
Source: Input
Output
Text (retrieved 2026-10-29)
Source: Output

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

Weights
Open weights
Source: Weights
Access
API, Local
Source: Access
License
Apache 2.0
Source: License
Shutdown
None announced
no source

API

API model ID
jamba-mini, currently jamba-mini-2-2026-01
Source: API model ID
API model ID
jamba-mini (retrieved 2026-10-29)
Source: API model ID

Further details

Notes
Mixture-of-Experts Jamba2 variant for reliable output with a lean memory footprint. The undated jamba-mini API endpoint points to jamba-mini-2-2026-01; Jamba Mini 1.7 was deprecated on February 1, 2026.
Source: Notes

Frequently asked questions about Jamba2 Mini

How much does Jamba2 Mini cost?

Through the API $0.20 per 1M input tokens and $0.40 output. That is cheap.

Prices from Nov 20, 2026

Output price since tracking beganDollars per 1M tokens · lower is better

Nov 1, 2026Nov 20, 2026

Unchanged since tracking began on Nov 1, 2026.

How large is the context window of Jamba2 Mini?

256,000 tokens. Separate limits may apply to input and output.

Context windowContext limit in tokens · higher is better
Kimi K2.6262K
Qwen 3.8 27B262K
Kimi K2.7 Code262K
Kimi K2.5262K
Kimi K2 Thinking262K
Mistral Medium 3.5262K
Jamba2 MiniRank 85 of 112256K

Can I run Jamba2 Mini locally?

Yes, the weights are open.

Stated by AI21 Labs

Open weights
yes
License
Apache 2.0
Parameters
52B
active per token
12B

Where can I use Jamba2 Mini?

Through the API. Subscription access is not documented in the available plan data.

Through the API

$0.20 input, $0.40 output per 1M tokens, unchanged since Nov 1, 2026.

Sources and data date

Show all 15 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

EvaluatorTest seriesComposite indicesRecorded active measurements
Vals AI101

Without vendor claims and historical or disputed results; those stay labelled in the single values.

TypeEvidence and data date
Price

Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs.

Input

Documented modalities. Retrieved October 29, 2026.

Output

Documented modalities. Retrieved October 29, 2026.

Research statusResearch date October 4, 2026. 9 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not published

evaluation-details

Note

Task snapshots, decoding parameters and score units are not specified consistently across the model-card table.

Additional sourceAI21 Jamba2 announcement (retrieved October 3, 2026; October 4, 2026) · AI21 Jamba2 announcement · Editorial description reviewed October 4, 2026
Additional sourceAI21 Jamba foundation models (retrieved October 29, 2026)
Additional sourceAI21 Jamba Mini 1.6 model-card table (retrieved October 15, 2026)
Additional sourceGitChameleon 2.0, ACL 2026 paper (retrieved October 17, 2026)
Additional sourceVals AI LiveCodeBench leaderboard (retrieved October 19, 2026)
Additional sourceAI21 Jamba 1.6 model-card table (retrieved October 22, 2026)
Additional sourceVals AI Jamba 1.6 Mini model page (retrieved October 24, 2026)
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0