Model components / Fine-tuned / chat models

K2-V2

LLM360 / MBZUAI's K2-V2 (70B dense), released Dec 2025: a 360-open, reasoning-enhanced LLM. Fully reproducible — Apache-2.0 weights plus the complete TxT360 pretraining, midtraining and SFT data, training code and checkpoints. Supersedes the 65B K2 Chat.

Apache-2.0; open weights + open TxT360 data + open training code + checkpoints. Outperforms Qwen2.5-72B and approaches Qwen3-235B; strongest fully-open model by parameter scale, though OLMo 3 leads on fully-open evals.

Openness

5 high confidence

5.0

weights: open(Apache-2.0)
data: open(TxT360 pretraining + midtraining + SFT)
code: open(k2v2_train)
checkpoints: open
license: Apache-2.0(OSI)

360-open / fully reproducible: Apache-2.0 weights + complete TxT360 data + training code + checkpoints. A clean fully-open exemplar at 70B.

https://huggingface.co/LLM360/K2-V2 recorded 2026-06-30
K2-V2 70B; Apache-2.0; open data/code/checkpoints
https://arxiv.org/abs/2512.06201 recorded 2026-06-30
K2-V2: A 360-Open, Reasoning-Enhanced LLM (Dec 2025); TxT360 data fully released

Adoption

2 medium confidence

2.0

New (Dec 2025) and still niche, like its K2 predecessor; included as a clean fully-open 70B exemplar rather than on popularity.

https://huggingface.co/LLM360/K2-V2 recorded 2026-06-30
K2-V2 repo live on HF; niche download volume

Capability

4 medium confidence

4.0

Strongest fully-open model by parameter scale (70B dense), with strong long-context and math/reasoning; OLMo 3 still leads on fully-open evals. Card/report-reported numbers -> medium confidence.

https://arxiv.org/abs/2512.06201 recorded 2026-06-30
outperforms Qwen2.5-72B, approaches Qwen3-235B

Unchanged since 2026-07-29 (last edited, not re-checked)