Tasks and inputs
Sage S2.1 Flash is listed for Image-text-to-text, Multimodal, Text generation. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
Sage S2.1 Flash is Shanghai AI Lab's open scientific multimodal agent model, reading molecules, crystals and proteins alongside text and images.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No API pricing stated; open weights for self-hosting
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Ask a general model about a crystal structure and you first have to turn the structure into text. Sage S2.1 Flash skips that step. It accepts more than ten kinds of scientific input, including molecular graphs, crystal structures, protein sequences and 3D protein structures, and reasons over them next to ordinary language.
Shanghai AI Laboratory released it in mid-November 2026 for research tasks that need several rounds of analysis and checking: it can run specialised software, compare intermediate outputs and visual feedback, and change course.
Each input type passes through its own encoder. A projector maps those features into the same space as the text token embeddings, and the combined sequence goes into the language model. This is the Sage SciPerception representation. In practice it means extra setup: the scientific encoders, task readouts and preprocessing are installed separately, and each task's input format is defined in the configuration.
Across fifteen benchmarks against GLM-5.3-Flash, DeepSeek-V4.1-Flash-1017, Qwen3.8-Max, GPT-5.6 Sol and Claude Opus 5, the lab says it placed first on four and second on seven. Highlights:
The coding gap is real. This is an office, ML-engineering and science specialist.
| Field | Value |
|---|---|
| Hub repo | internlm/Sage-S2.1-Flash (FP8 sibling posted 13 November) |
| Released | 15 November 2026 |
| Parameters | About 321B (MoE, 288 experts, 8 active, 45 layers) |
| Inputs | Text, images, 10+ scientific structure types |
| Licence | MIT |
It is the lighter, science-equipped counterpart to Sage S2.1, the 744B agent formerly called Atria Dawn. For hosted fast models in the same weight class, GLM 5.3 Flash and DeepSeek V4.1 Flash are the obvious comparisons.
Choose five property-prediction or structure-inference tasks from your own pipeline. Run them with native structure input, then again with the same structures serialised as text, and see whether the native encoders actually help. Separately, time one multi-step analysis that calls your real tools. Dedicated deployment guides cover coding, office, security and computer-use setups.
Compare the model on the work you need to complete.
Sage S2.1 Flash is listed for Image-text-to-text, Multimodal, Text generation. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under MIT. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
sensetime
SenseNova-Loop-9B is SenseTime's 9B looped multimodal model: 32 layers run as 56, 512K context, ~15 GiB at 256K in 4-bit, open weights under Apache 2.0.
Tencent
Tencent Hy4 Flash is the low-cost Hy4 tier with open weights: a 220B MoE (~11.3B active) for agents and documents, with 1M context and image input.
Meituan
LongCat-2.5 is Meituan's open-weight MIT-licensed multimodal MoE for long-horizon agents: screenshots, charts, video and multi-step coding.
sensetime
SenseNova-Loop-9B-Preview is SenseTime's looped 9B text-and-image model for long-context agentic work, running shared layers twice for extra depth.
Liquid AI
LFM3-VL-2B is Liquid AI's open 2B vision-language model for images and video on-device, strong on documents, charts and GUI screenshots.
NVIDIA
Nemotron 3.5 Super VL 120B A12B is NVIDIA's open multimodal reasoning MoE: 12B active, 1M context, image and video input, switchable thinking.