Language modelOpen source
Jamba Reasoning 3B
AI21
- Released
- October 8, 2025
- Data date
- October 3, 2026
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.
AI21 releases the 3B model under Apache 2.0 with a 256K context window. The provider documents local use through LM Studio and llama.cpp, among other options.
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.
IFBench (Show measurement, test conditions, and source)
- Source value
- 52
- Score
- 52
- Metric
- IFBench
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- AI21 announcement chart pixels
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Component chart against small-model peers; separate composite combines MMLU-Pro, LiveCodeBench, GPQA-Diamond, SciCode, IFBench and HLE at 32K context on Apple M3. Composite is not one benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort. AI21 self-reported chart; no exact independent record
Humanity's Last Exam (Show measurement, test conditions, and source)
- Source value
- 6
- Score
- 6
- Metric
- Humanity's Last Exam
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- AI21 announcement chart pixels
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Component chart against small-model peers; separate composite combines MMLU-Pro, LiveCodeBench, GPQA-Diamond, SciCode, IFBench and HLE at 32K context on Apple M3. Composite is not one benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort. AI21 self-reported chart; no exact independent record
MMLU-Pro (Show measurement, test conditions, and source)
- Source value
- 61
- Score
- 61
- Metric
- MMLU-Pro
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- AI21 announcement chart pixels
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Component chart against small-model peers; separate composite combines MMLU-Pro, LiveCodeBench, GPQA-Diamond, SciCode, IFBench and HLE at 32K context on Apple M3. Composite is not one benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort. AI21 self-reported chart; no exact independent record
Sources and data date
Every statement links to its underlying documentation or leaderboard.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Additional source | AI21 Jamba Reasoning 3B announcement (retrieved October 3, 2026; October 4, 2026) · AI21 Jamba Reasoning 3B announcement · Editorial description reviewed October 4, 2026 |