ByteDance SeedLanguage modelWeights available
Cola2
Cola2 has no overall index value yet because there are too few independent benchmark results.
Summary
Cola2 is a research language model that divides text into spans and encodes them as compact continuous representations. A diffusion model processes those representations before a decoder generates text.
ByteDance Seed published the checkpoint on November 6, 2026. Weights and local inference code are available. The model card assigns Apache-2.0 to the repository code, with separate terms for weights and other assets.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 9 published benchmark results
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 (Show measurement, test conditions, and source)
- Source value
- 69.91
- Score
- 69.91
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
NarrativeQA (Show measurement, test conditions, and source)
- Source value
- 32.7
- Score
- 32.7
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
HellaSwag (Show measurement, test conditions, and source)
- Source value
- 41.75
- Score
- 41.75
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
StoryCloze (Show measurement, test conditions, and source)
- Source value
- 80.42
- Score
- 80.42
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
LAMBADA (Show measurement, test conditions, and source)
- Source value
- 63
- Score
- 63
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
ARC-Easy (Show measurement, test conditions, and source)
- Source value
- 87.62
- Score
- 87.62
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
TruthfulQA (Show measurement, test conditions, and source)
- Source value
- 54.1
- Score
- 54.1
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
MMLU (Show measurement, test conditions, and source)
- Source value
- 51.32
- Score
- 51.32
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
Cola2 (8-task average) (Show measurement, test conditions, and source)
- Source value
- 60.1
- Score
- 60.1
- Metric
- accuracy
- Unit
- %
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- ByteDance Seed Cola2 post-pretraining evaluation
- Status
- active
- Retrieved at
- 2026-11-07
- Methodology
- Vendor-reported result. Published Cola2 checkpoint after 346.03 billion source tokens. Three independently sampled latent-noise runs per task, eight DiT steps, DiT guidance 10, decoder guidance -0.9 and greedy token selection.
Similar models
More models of the same kind as Cola2, chosen by model line, developer, size, price and release date.
- Seed-OSS-36B-InstructByteDance Seed · same developer
- Nemotron 3.5 Bolt E4BNVIDIA · similar size
- MiMo V2.6 Distill Qwen 9BXiaomi · similar size
- Ornith 1.0 9BDeepReinforce · similar size
- Qwen 3.5 9BAlibaba · similar size
- Ministral 3 8BMistral AI · similar size
- Qwen 3 VL 8BAlibaba · similar size
- Rnj-1 InstructEssential AI · similar size
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
- License
- Apache-2.0 applies to the repository code. The model card says the weights have separate terms and gives no license identifier for them.
- Source: License
- Shutdown
- None announced
- no source
Frequently asked questions about Cola2
Can I run Cola2 locally?
Yes, the weights are open.
Stated by ByteDance Seed
- Open weights
- yes
- License
- Apache-2.0 applies to the repository code. The model card says the weights have separate terms and gives no license identifier for them.
- Parameters
- 8.8B
- active per token
- n/a
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 November 7, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Additional source | ByteDance Seed Cola2 model card (retrieved November 7, 2026) · ByteDance Seed Cola2 model card · Editorial description reviewed November 7, 2026 |
| Additional source | Cola2 weight release commit (retrieved November 7, 2026) · Cola2 weight release commit · Editorial description reviewed November 7, 2026 |
| Additional source | Cola2 Safetensors metadata (retrieved November 7, 2026) |
| Additional source | Cola2 inference repository (retrieved November 7, 2026) |
| Additional source | ByteDance Seed Cola2 post-pretraining evaluation (retrieved November 7, 2026) |