InternLM

★ 4/5 · Open Source Model

Shanghai AI Laboratory's Intern family publishes open-weight language, multimodal and scientific models under Apache 2.0, running from the compact InternLM3-8B-Instruct up to the 397B-parameter Intern-S2 series. Alongside the weights it maintains a full open toolchain covering pretraining, fine-tuning, deployment and evaluation.

Pros

Code and weights are Apache 2.0 with free commercial use stated on the lab's own site, and the Hugging Face org carries 124 models and 26 datasets rather than a single headline release
InternLM3-8B-Instruct was trained on only 4T tokens, cutting training cost by over 75%, yet scores 83.1 CMMLU, 76.6 MMLU and 51.1 on AlpacaEval 2.0 length-controlled win rate against Qwen2.5-7B and Llama3.1-8B, while InternLM2.5-7B-Chat-1M handles a one-million-token context
the surrounding toolchain is the real draw, with XTuner for fine-tuning, LMDeploy for serving, OpenCompass for evaluation and Lagent for agents all open sourced, and every model loads unmodified in Transformers, vLLM, SGLang, LMDeploy and Ollama.

Cons

The mainline chat series has had no new base model since InternLM3-8B-Instruct in January 2025, with momentum shifting to the Intern-S scientific line, so anyone expecting a frontier general chat model will be disappointed
the hosted API console and chat app sit on a Chinese-language site behind a sign-in, which pushes most non-Chinese users to self-host by default
the largest releases, Intern-S2-Preview-397B and the 241B Intern-S1, will not fit on a single GPU and no international inference partner offers them as a managed endpoint
English documentation and community support are noticeably thinner than Qwen's or DeepSeek's.

Cost

Free. The code and weights are released under Apache 2.0 with commercial use permitted, and the site carries a commercial-licence request form rather than a price. There is no USD rate card because the models are self-hosted, and the hosted intern-ai API console, with its own token allowances and rate limits, sits behind a Chinese-language sign-in with no published pricing.

Verdict

Best for research teams and self-hosters who want permissively licensed bilingual Chinese-English models plus a complete open training, deployment and evaluation stack, rather than anyone shopping for a managed API with predictable per-token billing.

Visit InternLM