InternLM has published Intern-S2-397B, a multimodal foundation mode... - Vibeus
InternLM has published Intern-S2-397B, a multimodal foundation model aimed at scientific intelligence and long-horizon agents. Its official model card says the model combines a visual pre-training approach with large-scale multitask reinforcement learning and agent reinforcement learning across more than 20 scientific domains. Thinking mode is enabled by default; the card lists maximum inference lengths of 256K tokens for text reasoning and 64K for multimodal benchmarks. The model can be deployed with LMDeploy, vLLM or SGLang, but its listed 403B-parameter size makes this a large self-hosting release rather than a lightweight local model. Источник