inclusionAILanguage modelActive
Ling 3.1 Flash Base
Ling 3.1 Flash Base has no overall index value yet because there are too few independent benchmark results.
Summary
Ling 3.1 Flash Base is the base checkpoint in the Ling 3.1 Flash series. inclusionAI recommends it for further training, adaptation, and research, not as a ready-to-use chat assistant. The model card lists 560 billion total parameters and 25 billion active parameters per token. The released weights are BF16 and come from FP8 training followed by Warmup-Stable-and-Merge checkpoint merging.
The model card contains 47 results across eight domains. It lists the metric and shot configuration for each row. These are vendor reports without published sample counts or confidence intervals. They belong only to the base checkpoint and are not attributed to the upgraded 1019 checkpoint or the existing chat model.
The configuration sets max_position_embeddings to 262,144. The model card also reports RULER results through 256K. These are separate figures and are not treated as the same measurement.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 47 published benchmark results
Page 1 of 4
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.
Race-High (Show measurement, test conditions, and source)
- Source value
- 93.88
- Score
- 93.88
- Metric
- EM
- Unit
- %
- Category
- Language understanding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 0-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
SquadV2 (Show measurement, test conditions, and source)
- Source value
- 92.79
- Score
- 92.79
- Metric
- Acc
- Unit
- %
- Category
- Language understanding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 1-shot; metric Acc; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
Belebele (Show measurement, test conditions, and source)
- Source value
- 94.67
- Score
- 94.67
- Metric
- EM
- Unit
- %
- Category
- Language understanding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 0-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
C3 (Show measurement, test conditions, and source)
- Source value
- 83.34
- Score
- 83.34
- Metric
- EM
- Unit
- %
- Category
- Language understanding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 0-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
CCPM (Show measurement, test conditions, and source)
- Source value
- 91.88
- Score
- 91.88
- Metric
- EM
- Unit
- %
- Category
- Language understanding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 0-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
ARC-c (Show measurement, test conditions, and source)
- Source value
- 97.97
- Score
- 97.97
- Metric
- EM
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 0-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
AGIEval (Show measurement, test conditions, and source)
- Source value
- 82.59
- Score
- 82.59
- Metric
- Acc
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. few-shot (0-5); metric Acc; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
SimpleQA-Verified (Show measurement, test conditions, and source)
- Source value
- 52.4
- Score
- 52.4
- Metric
- Acc
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 5-shot; metric Acc; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
C-SimpleQA (Show measurement, test conditions, and source)
- Source value
- 77.43
- Score
- 77.43
- Metric
- Acc
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 5-shot; metric Acc; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
AA-Omniscience (Show measurement, test conditions, and source)
- Source value
- 37.67
- Score
- 37.67
- Metric
- Acc
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 3-shot; metric Acc; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
MMLU-Pro (Show measurement, test conditions, and source)
- Source value
- 76.71
- Score
- 76.71
- Metric
- EM
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 5-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
CEval (Show measurement, test conditions, and source)
- Source value
- 92.36
- Score
- 92.36
- Metric
- EM
- Unit
- %
- Category
- Knowledge
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- inclusionAI Ling-3.1-flash-base model card, Base Model Evaluation table
- Status
- active
- Retrieved at
- 2026-10-29
- Methodology
- Vendor-reported result. 5-shot; metric EM; score as listed for Ling-3.1-flash-base in the official 47-benchmark base evaluation table.
- Context
- Vendor-reported result from the Ling-3.1-flash-base model card; no sample count or confidence interval is published. The 47-result table belongs to the base checkpoint only.
Page 1 of 4
All specifications
Architecture
- Architecture
- Hybrid-linear MoE with 63 KDA and 9 Gated MLA layers, 512 routed experts, one shared expert, and eight active routed experts per token.
- Source: Architecture
- Architecture
- Hybrid-linear MoE with 63 KDA and 9 Gated MLA layers, 512 routed experts, one shared expert, and eight active routed experts
- Source: Architecture
Licence and access
Further details
- Checkpoint type
- Base checkpoint for continued pretraining, mid-training, fine-tuning, and research; the model card does not recommend direct end-user chat without additional post-training.
- Source: Checkpoint type
- Parameter scale
- 560 billion total parameters, with 25 billion active per token.
- Source: Parameter scale
- Training and weight precision
- FP8 training; the released base files are the BF16 checkpoint after Warmup-Stable-and-Merge checkpoint merging.
- Source: Training and weight precision
- Configuration position limit
- config.json sets max_position_embeddings to 262,144. The benchmark table reports RULER results through ruler_256k; that is a separate observed benchmark length.
- Source: Configuration position limit
- Benchmarks
- The model card reports 47 scores across language understanding, knowledge, code, mathematics, reasoning, multilingual ability, medicine, and long context. Each table row lists its shot count and metric.
- Source: Benchmarks
- Training and weight precision
- FP8 training; BF16 weights are provided after Warmup-Stable-and-Merge
- Source: Training and weight precision
- Context configuration
- config.json sets max_position_embeddings to 262,144; the RULER table includes results through 256k
- Source: Context configuration
- Benchmark scope
- 47 benchmark results across eight domains with metrics and shot configurations
- Source: Benchmark scope
Frequently asked questions about Ling 3.1 Flash Base
How large is the context window of Ling 3.1 Flash Base?
262,144 tokens. That is how much text Ling 3.1 Flash Base reads at once, about 390 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Ling 3.1 Flash BaseRank 68 of 100262K
Kimi K2.6262K
Qwen 3.8 27B262K
Kimi K2.7 Code262K
Kimi K2.5262K
Kimi K2 Thinking262K
Ling 3.0 Flash262K
Can I run Ling 3.1 Flash Base locally?
Yes, the weights are open. With 560B parameters it needs a lot of memory, more a server than a laptop.
Stated by inclusionAI
- Open weights
- yes
- License
- MIT
- Parameters
- 560B
- active per token
- 25B
Sources and data date
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 |
|---|---|
| Research status | Research date October 29, 2026. 58 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | Benchmark sample counts and confidence intervals Note The 47-row table lists metric, shot configuration, and score, but no per-benchmark n or confidence intervals. |
| Not found in the inspected sources | Public weight availability date Note No verified public release date was found in the card and config. The repository creation timestamp is not treated as a release date. |
| 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 | Ling-3.1-flash-base official model card (retrieved October 29, 2026) · Ling-3.1-flash-base official model card · Editorial description reviewed October 29, 2026 |
| Additional source | Ling-3.1-flash-base configuration (retrieved October 29, 2026) · Ling-3.1-flash-base configuration · Editorial description reviewed October 29, 2026 |
| Additional source | Model metadata (retrieved October 29, 2026)12 |