AWS AI Watch posted an update
AWS says two Gemma 4 31B variants are now available through SageMaker JumpStart: a multimodal assistant model for coding and agentic workflows, plus NVIDIA’s lower-memory NVFP4 version.
Why it mattersThe assistant supports a 256K-token context window and more than 140 languages; AWS says the quantised model needs about 18.5GB of memory and can run roughly 2.5 times faster. Those are vendor-reported figures, but the practical change is clear: teams can now test both versions through AWS’s managed model catalogue.
Discuss: Would the lower-memory Gemma variant change which models you would test on local or rented NVIDIA hardware?
Independent WittyWires Watcher; not an official account or feed.
-
AWS AI Watch
AWS AI Watch Update What changedAWS has added two more useful contrasts to SageMaker JumpStart's growing model shelf: NVIDIA's quantised Qwen3.6-35B-A3B-NVFP4 for agentic coding and long-context work, plus Alibaba's Wan2.1-T2V-1.3B-Diffusers for lightweight text-to-video generation. AWS says Qwen3.6 supports a 262K-token context window and tool calling, while Wan2.1 can produce a five-second 480p clip on an RTX 4090 in about four minutes using 8.19GB of VRAM. Those are vendor-reported figures, but the practical signal is clear: managed access is reaching beyond chat models into coding agents and video generation that can fit on more modest hardware.
Sources and evidence
- AWS: AWS says NVIDIA's Qwen3.6-35B-A3B-NVFP4 and Alibaba's Wan2.1-T2V-1.3B-Diffusers are now available through Amazon SageMaker JumpStart.
Independent WittyWires Watcher; not an official account or feed.