Model components / Base / pretrained models

InternLM 2.5

Shanghai AI Lab's open-weights model family. The InternLM 2.5 generation (7B/20B, July 2024, 1M context via dynamic NTK interpolation) was superseded by InternLM3 in January 2025, an 8B model trained on only 4T tokens that reportedly matches GPT-4o-mini, with integrated 'deep thinking' chain-of-thought mode and ~4x the data efficiency (IQPT) of Llama 3.1. A leading Chinese academic-origin foundation model, heavily used in the Chinese research community.

InternLM2.5-7B base (~8B params, 1M-token context), released 2024 (InternLM2 tech report arXiv:2403.17297). Base SKU scored here; InternLM2.5-7B-Chat is the instruct SKU (finetuned_chat). Verified live on HF June 2026; a 2024-era 7B model.

Openness

3 medium confidence

3.0

weights: open(custom 'Free Commercial License', academic free, commercial use free but requires an application form)
data: closed(pretraining corpus not released)
code: open(Apache-2.0 training/inference code on GitHub)
license: Apache-2.0(code, OSI) + custom weights license(non-OSI, application step)

Code is OSI (Apache-2.0) but the weights license is a custom non-OSI 'Free Commercial License' requiring a commercial-license application form (HF card confirms form link, 2026-06-04), so open_weights not open_source; flagged for the non-OSI weights term.

https://huggingface.co/internlm/internlm2_5-7b recorded 2026-06-04
model card: code Apache-2.0; weights fully open for academic research and free commercial usage via application form (QQ survey link)
https://github.com/InternLM/InternLM/blob/main/model_cards/internlm2.5_7b.md recorded 2026-06-04
InternLM2.5-7B base model card with benchmark results

Adoption

2 medium confidence

2.0

HF monthly downloads modest in 2026 (3,521/mo for the 7B base, confirmed 2026-06-04) as a 2024-era model; distributed via HF, Ollama, ModelScope. Cumulative reach across the InternLM2.5 family + mirrors sits in the early-adopter band; specialist/research uptake, not production scale.

https://huggingface.co/internlm/internlm2_5-7b recorded 2026-06-04
3,521 downloads last month
https://ollama.com/internlm/internlm2.5 recorded 2026-06-04
InternLM2.5 distributed via Ollama

Capability

2 medium confidence

2.0

Strong sub-8B model for 2024 with 1M-context, but below 2026 frontier; calibrates near Falcon 3 (C2) / Yi-1.5 (C2) on the dated small-model ladder. MMLU 71.6 confirmed on HF card 2026-06-04.

https://huggingface.co/internlm/internlm2_5-7b recorded 2026-06-04
InternLM2.5-7B base MMLU 71.6 (OpenCompass)
https://github.com/InternLM/InternLM/blob/main/model_cards/internlm2.5_7b.md recorded 2026-06-04
InternLM2.5-7B base MMLU 71.6 (OpenCompass)

Unchanged since 2026-07-30 (last edited, not re-checked)