InclusionAILanguage modelOpen source
Ling-3.0-tiny
Ling-3.0-tiny has no overall index value yet because there are too few independent benchmark results. From the vendor-reported results we estimate about 56.0 points (95% range 36.5 to 75.6). This is not an independent measurement.
45out of 100same as last month
Summary
Ling-3.0-tiny brings InclusionAI’s hybrid attention architecture to a 7.9B checkpoint with 1.3 billion active parameters. Three Kimi Delta Attention layers alternate with one Multi-Head Latent Attention layer. The provider documents local tests on DGX Spark and Apple Silicon Macs.
InclusionAI releases BF16, FP8, and INT4 weights under MIT. The model handles 262,144 context tokens and is also available through the OpenRouter route linked in its model card.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 30 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.
TAU3-Banking-AA (Show measurement, test conditions, and source)
- Source value
- 20.8
- Score
- 20.8
- Metric
- TAU3-Banking-AA
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
BFCL-v4 (FC) (Show measurement, test conditions, and source)
- Source value
- 62.72
- Score
- 62.72
- Metric
- BFCL-v4 (FC)
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
Terminal-Bench 2.1 (Show measurement, test conditions, and source)
- Source value
- 27.7
- Score
- 27.7
- Metric
- Terminal-Bench 2.1
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
ArtifactsBench (Show measurement, test conditions, and source)
- Source value
- 47.93
- Score
- 47.93
- Metric
- ArtifactsBench
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
SciCode (Show measurement, test conditions, and source)
- Source value
- 24.2
- Score
- 24.2
- Metric
- SciCode
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
GPQA Diamond (Show measurement, test conditions, and source)
- Source value
- 73.4
- Score
- 73.4
- Metric
- GPQA Diamond
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
HLE (Show measurement, test conditions, and source)
- Source value
- 9.3
- Score
- 9.3
- Metric
- HLE
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
IFBench (Show measurement, test conditions, and source)
- Source value
- 63.61
- Score
- 63.61
- Metric
- IFBench
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
Multi-IF (Show measurement, test conditions, and source)
- Source value
- 83.15
- Score
- 83.15
- Metric
- Multi-IF
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- HF model-card table reproducing Artificial Analysis
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Terminal-Bench 2.1 uses AA protocol, Terminus 2 default harness, 2-hour timeout, JSON parser, preserve-thinking, 3 runs/task mean, temperature 1.0, max_new_tokens 32K, 256K context.
- Context
- Source-specific observation; it is not a shared comparison cohort. AA-protocol/provider-card evidence, not an exact committed independent record
AA-Briefcase v1.1 rubric pass rate (Show measurement, test conditions, and source)
- Source value
- 8.383838383838384
- Score
- 8.383838383838384
- Metric
- share of binary rubric checks passed
- Unit
- %
- Benchmark version
- v1.1
- Category
- agentic
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis, Ling 3.0 Tiny model page
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. AA-Briefcase v1.1, one run over 91 tasks across four scenarios. This is the share of binary rubric checks passed across tasks, not the share of tasks fully passed. The per-profile rubric-check denominator and reasoning effort are not disclosed.
- Context
- The AA profile page does not expose an exact model-side reasoning effort or immutable serving revision. Its weights source is Ling-3.0-tiny-int4. This is a dated named-route observation, not a pinned full-precision checkpoint or a shared comparison cohort.
AutomationBench-AA score (Show measurement, test conditions, and source)
- Source value
- 0.2807334937855976
- Score
- 0.2807334937855976
- Metric
- share of task objectives completed with no guardrail violations
- Unit
- %
- Benchmark version
- AA current implementation
- Category
- agentic
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis, Ling 3.0 Tiny model page
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. AutomationBench-AA evaluates 657 simulated SaaS workflow tasks with REST API tools, one repeat per AA methodology. Its headline Score is the share of task objectives completed with no guardrail violations; this is distinct from the separate Tasks Completed metric. Per-profile scored-objective denominator and model reasoning effort are not disclosed.
- Context
- The AA profile page does not expose an exact model-side reasoning effort or immutable serving revision. Its weights source is Ling-3.0-tiny-int4. This is a dated named-route observation, not a pinned full-precision checkpoint or a shared comparison cohort.
Terminal-Bench 4.0 (Show measurement, test conditions, and source)
- Source value
- 0
- Score
- 0
- Metric
- pass@1 task success rate
- Unit
- %
- Benchmark version
- 4.0
- Category
- coding
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis, Ling 3.0 Tiny model page
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Artificial Analysis Terminal-Bench 4.0: 66 terminal-execution tasks with three repeats in the published methodology, scored by test-suite pass@1. The source does not state this profile's actual completed-run count, exact harness settings or reasoning effort.
- Context
- The AA profile page does not expose an exact model-side reasoning effort or immutable serving revision. Its weights source is Ling-3.0-tiny-int4. This is a dated named-route observation, not a pinned full-precision checkpoint or a shared comparison cohort.
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 5, 2026
Ling-3.0-tinyK2 Horizon MoVA 36B-A4BClaude Fable 5ERNIE 4.5 21B A3BLlama 4 Maverick
All specifications
Architecture
Input and 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
- Access
- Local weights; the model card links inclusionai/ling-3.0-tiny:free on OpenRouter
- Source: Access
Frequently asked questions about Ling-3.0-tiny
How good is Ling-3.0-tiny in benchmarks?
Independent results are not yet enough for the overall index. An estimate from 3 vendor-reported results (GPQA Diamond 73.4%, Humanity's Last Exam (no tools) 9.3%, Terminal-Bench 2.1 27.7%) gives about 56.0 points, with a 95% range of 36.5 to 75.6. The estimate is calibrated on models with both vendor-reported and independent results. Vendor numbers run clearly above independent measurements on some tests; the calibration offsets that on average.
Vendor estimate, not an overall index value
How large is the context window of Ling-3.0-tiny?
262,144 tokens. That is how much text Ling-3.0-tiny reads at once, about 390 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Can I run Ling-3.0-tiny locally?
Yes, the weights are open.
Stated by InclusionAI
- Open weights
- yes
- License
- MIT
- Parameters
- 7.9B
- active per token
- 1.3B
Sources and data date
Show all 19 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
| Evaluator | Test series | Composite indices | Recorded active measurements |
|---|---|---|---|
| Artificial Analysis | 6 | 1 | 12 |
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 4, 2026. 11 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | CritPt current score Note The AA exact model page exposes CritPt=0, but the official CritPt leaderboard checked 2026-10-17 is explicitly marked Under review and says its authors are auditing the benchmark after external feedback. No CritPt result is added as a confirmed observation pending the reviewed dataset and reruns. |
| Unresolved | exact serving revision, reasoning effort and per-metric executed N Note The AA profile identifies a named current route and lists Ling-3.0-tiny-int4 as weights source, but publishes no immutable served revision or exact contestant reasoning effort. Its method pages disclose benchmark protocols, not actual completed per-profile N for most metrics. No such values are inferred from task plans or repetitions. |
| Not published | score scale, denominator, serving revision and confidence interval Note The newly retained table cells do not publish their score scale, denominator, serving revision or 95% confidence interval. |
| Additional source | Ling-3.0-tiny model card (retrieved October 3, 2026; October 4, 2026; October 17, 2026; October 19, 2026) · Ling-3.0-tiny model card · Editorial description reviewed October 4, 2026 |
| Additional source | Ling-3.0-flash-VL research card (retrieved October 3, 2026) · Ling-3.0-flash-VL research card · Editorial description reviewed October 4, 2026 |
| Additional source | Ling-3.0-flash-Fin research card (retrieved October 3, 2026) · Ling-3.0-flash-Fin research card · Editorial description reviewed October 4, 2026 |
| Additional source | Artificial Analysis, Ling 3.0 Tiny model page (retrieved October 17, 2026) |
| Additional source | Ardana, ling-3.0 benchmarks (retrieved October 19, 2026) |
| Additional source | Model metadata (retrieved October 17, 2026)12345678 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |