Liquid AILanguage modelOpen weights
LFM3-VL-2B
LFM3-VL-2B has no overall index value yet because there are too few independent benchmark results.
Summary
LFM3-VL-2B accepts text, native-resolution images, and video. The dense model has 1.92 billion parameters, a 128K context window, and an in-house vision encoder. Its card publishes 62 visual benchmarks with thinking enabled and disabled.
The model card separates individual tests, category means, and the overall average. Edge and B300 measurements, plus results for a separately fine-tuned manipulation VLA, apply only to their documented configurations. Robotics results are not scores for unmodified LFM3-VL-2B.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 161 published benchmark results
Page 1 of 14
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.
MME (Show measurement, test conditions, and source)
- Source value
- 86.9
- Score
- 86.9
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking enabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MME (Show measurement, test conditions, and source)
- Source value
- 76.7
- Score
- 76.7
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking disabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MMStar (Show measurement, test conditions, and source)
- Source value
- 74.4
- Score
- 74.4
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking enabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MMStar (Show measurement, test conditions, and source)
- Source value
- 62.2
- Score
- 62.2
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking disabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
SimpleVQA (Show measurement, test conditions, and source)
- Source value
- 36.3
- Score
- 36.3
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking enabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
SimpleVQA (Show measurement, test conditions, and source)
- Source value
- 27.7
- Score
- 27.7
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking disabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
SEED-Bench (image) (Show measurement, test conditions, and source)
- Source value
- 78.6
- Score
- 78.6
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking enabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
SEED-Bench (image) (Show measurement, test conditions, and source)
- Source value
- 76.8
- Score
- 76.8
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking disabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MMBench (dev EN v1.1) (Show measurement, test conditions, and source)
- Source value
- 86.1
- Score
- 86.1
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking enabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MMBench (dev EN v1.1) (Show measurement, test conditions, and source)
- Source value
- 76.5
- Score
- 76.5
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking disabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MMMU (val) (Show measurement, test conditions, and source)
- Source value
- 55.7
- Score
- 55.7
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking enabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
MMMU (val) (Show measurement, test conditions, and source)
- Source value
- 48.6
- Score
- 48.6
- Metric
- percent
- Unit
- %
- Category
- General
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- LiquidAI/LFM3-VL-2B official model card at revision ba44bd51db20a2753f28c273a1571ccaa210d954
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. Published LFM3-VL-2B Thinking disabled setting; vLLM 0.27.1, recommended generation parameters, normalized score 0 to 100.
- Context
- Manufacturer-reported. Thinking results belong only to the stated configuration and do not imply results for other model modes.
Page 1 of 14
All specifications
Architecture
- Architecture
- 1.73 billion language-model parameters, 164 million in the vision encoder, and 25 million in the projector. The language backbone has 24 layers; the vision encoder has 14.
- Source: Architecture
Input and output
- Languages
- 16 listed text languages: English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, and Polish.
- Source: Languages
Licence and access
- Access
- Open weights, local deployment, and the API and Playground linked by the publisher
- Source: Access
Further details
- Context and video
- 128K context tokens. The card lists a default of 2 fps and 4 to 768 frames for video; the chosen frame count affects context and memory use.
- Source: Context and video
- Thinking configurations
- The model card publishes benchmark tables with thinking enabled and disabled. Results with thinking enabled and disabled are separate evaluations.
- Source: Thinking configurations
Frequently asked questions about LFM3-VL-2B
How large is the context window of LFM3-VL-2B?
131,072 tokens. That is how much text LFM3-VL-2B reads at once, about 200 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Can I run LFM3-VL-2B locally?
Yes, the weights are open.
Stated by Liquid AI
- Open weights
- yes
- License
- LFM Open License v1.0. See the license text for its terms.
- Parameters
- 1.9B
- active per token
- n/a
Sources and data date
Show all 9 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 29, 2026. |
| Output | Documented modalities. Retrieved October 29, 2026. |
| Research status | Research date October 29, 2026. 62 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | GPU throughput chart raw points Note The concurrency curve is published only as a graphic. Apart from the manufacturer’s “over 12,000 tokens/s” claim, machine-readable individual points are unavailable. |
| Unresolved | Robotics variant model IDs Note The article reports a fine-tuned manipulation VLA but does not provide a separate model name or exact checkpoint ID in this source. Its results therefore remain in a clearly separate configuration. |
| Not published | Sample counts for the 62 benchmarks Note Per-suite sample counts for the 62 benchmarks are mostly not stated; selected methodological details appear in the model card. |
| Not found in the inspected sources | Independent evaluation Note The reviewed sources do not link an independent reproduction of the LFM3-VL results. |
| Not found in the inspected sources | Technical report or research paper Note Reviewed publisher sources do not link a separate technical report or research paper for LFM3-VL-2B. |
| Not found in the inspected sources | Vals rows in captured payloads Note Nine complete Vals payloads dated 28 October were checked; no exact model ID or unambiguous model label. Limited to these saved benchmarks. |
| Unresolved | Uncaptured Arena Code/Speech leaderboards Note Text, Agent including Code/Chat/Work, and twelve WebDev views were captured. A separate general Code and Arena Speech leaderboard were not saved as complete tables. |
| Unresolved | Artificial Analysis catalog reconciliation Note The full current AA model and benchmark catalog was not re-reconciled for this profile in this sidecar. This sidecar makes no AA absence claim; native AA imports remain authoritative. |
| Additional source | Liquid AI LFM3 release announcement (retrieved October 29, 2026) · Liquid AI LFM3 release announcement · Editorial description reviewed October 29, 2026 |
| Additional source | Liquid AI model documentation (retrieved October 29, 2026) · LFM3 vision documentation · Editorial description reviewed October 29, 2026 |
| Additional source | Liquid AI LFM3 Hugging Face collection (retrieved October 29, 2026) |
| Additional source | Hugging Face model metadata (retrieved October 29, 2026) |
| Additional source | Liquid AI LFM3-VL-2B edge-inference chart (retrieved October 29, 2026) |
| Additional source | Model metadata (retrieved October 29, 2026)123 |