MultimodalFlow
← Back to Blog

Can DeepSeek's New V4-Flash-0731 Run on Jetson Thor? The 284B MoE Edge Boundary

DeepSeek-V4DeepSeek-V4-FlashJetson ThorEdge AIMoESGLangvLLMLocal Inference

After checking Qwen3.8-Max, I looked at DeepSeek's recent open-weight releases. The model to watch is DeepSeek-V4-Flash-0731, not the older DeepSeek-V3 or R1 releases.

Short answer:

QuestionAnswer
Recent DeepSeek open-weight model to watchDeepSeek-V4-Flash-0731
LicenseMIT
ScaleV4-Flash: 284B total / 13B activated
Context1M tokens
Can one Jetson Thor run it locally?Not realistically
Practical test pathRun V4-Flash on a multi-GPU server; use Thor for 30B/35B local edge comparisons

DeepSeek-V4-Flash-0731 is open and strong, but it is not a single-Jetson-Thor local model.


What Was Released

DeepSeek's Hugging Face organization shows deepseek-ai/DeepSeek-V4-Flash-0731 as a recent V4 release. The model card describes the V4-Flash family as:

ItemDeepSeek-V4-Flash-0731
ArchitectureMoE
Parameters284B total / 13B activated
Hugging Face model page size304B params
Context length1M tokens
PrecisionFP4 + FP8 mixed
LicenseMIT
Focuscode agents, tool use, long context, reasoning effort control

DeepSeek-V4 also includes the larger DeepSeek-V4-Pro, listed as 1.6T total / 49B activated with the same 1M context target. Pro is the ceiling; Flash is the more efficient variant.

The 0731 checkpoint has major agent benchmark improvements over the preview release:

BenchmarkV4-Flash-0731V4-Flash Preview
Terminal Bench 2.182.761.8
NL2Repo54.239.4
Cybergym76.738.7
DeepSWE54.47.3
Toolathlon-Verified70.349.7
Agents' Last Exam25.215.8

This is clearly aimed at coding agents, tool use, and long-context work rather than simple chat.


Thor Check

I checked two live Thor machines:

DeviceEnvironment
thorNVIDIA Thor, CUDA 13.0, Linux 6.8.12-tegra, 122 GiB memory
thor-jdNVIDIA Thor, CUDA 13.0, Linux 6.8.12-tegra, 122 GiB memory

Neither machine has DeepSeek or DSpark weights under /home/nvidia/models:

find /home/nvidia/models -maxdepth 2 -iname "*deepseek*" -o -iname "*dspark*"

So there is no local V4-Flash benchmark to report. More importantly, the model scale makes a single-Thor run the wrong target.


Why V4-Flash Does Not Fit One Thor

The phrase "13B activated" is easy to misread. It does not mean V4-Flash has the memory footprint of a 13B dense model.

V4-Flash is still a 284B-class MoE model. Even with FP4 experts and FP8 non-expert weights, it is far beyond Thor's comfortable 123 GB unified-memory envelope.

Rough weight-only estimates:

AssumptionSize
284B at FP8~284 GB
284B at INT4~142 GB
FP4 + FP8 mixeddepends on non-expert ratio, but still near or above Thor's limit
1M context KV cacheadds substantial memory pressure

That excludes CUDA workspace, MoE routing buffers, prefill tensors, speculative decoding state, and server overhead.

The official model card's serving examples point to this hardware class. The vLLM example uses a single 4×GB300 node with data parallelism, expert parallelism, FP8 KV cache, and DSpark speculative decoding. The SGLang example uses --tp 4 and a specialized MoE backend:

sglang serve \
  --model-path deepseek-ai/DeepSeek-V4-Flash-0731 \
  --tp 4 \
  --moe-runner-backend flashinfer_mxfp4 \
  --speculative-algorithm DSPARK

That is not a Jetson Thor deployment shape.


What Thor Should Test Instead

Use two layers:

LayerTest
Multi-GPU serverDeepSeek-V4-Flash-0731 agent quality
Jetson Thor30B/35B local edge models as a deployment baseline

For V4-Flash, measure real agent tasks:

MetricMeaning
Pass@1Does it solve the task on the first attempt?
Total wall timeHow long does the agent loop take?
Output tokensHow expensive is reasoning effort?
Tool-call countIs it efficient or wandering?
Failure modecommand error, hallucination, format failure, missing verification

For Thor, test models that actually fit:

ModelUse case
Qwen3-Coder-30B-A3B-FP8local coding agent
Qwen3.6-35B-A3B-FP8local general agent / long-context assistant
Qwable-v1Claude-style agent behavior distillation
DeepSeek-R1-Distill-Qwen-14B / 32Breasoning model if downloaded later

This comparison answers the useful engineering question: how far is a local 30B/35B Thor agent from a frontier open-weight DeepSeek V4 agent?


Verdict

DeepSeek-V4-Flash-0731 is one of the most interesting recent open-weight models: MIT License, 284B MoE, 13B activated, 1M context, and much stronger agent benchmarks than the preview.

But it is still too large for a single Jetson Thor. The official examples already point toward TP4, expert parallelism, and high-end multi-GPU servers.

My practical choice:

  • Test DeepSeek-V4-Flash-0731 on a multi-GPU server or hosted endpoint.
  • Use Thor for Qwen3-Coder-30B-A3B-FP8, Qwen3.6-35B-A3B-FP8, Qwable-v1, or DeepSeek-R1 distilled models.
  • Do not publish fake Thor token/s numbers for a model that is not locally present and not realistically supported by the hardware.

DeepSeek-V4-Flash-0731 is worth tracking, but Jetson Thor's role is a 30B/35B local edge baseline, not a 284B MoE single-node server.


Sources

  • DeepSeek Hugging Face organization: https://huggingface.co/deepseek-ai
  • DeepSeek-V4-Flash-0731 model card: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
  • DeepSeek-V4-Pro model card: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro
  • DeepSeek-V3 GitHub repository: https://github.com/deepseek-ai/DeepSeek-V3

Thor environment checked on August 13, 2026. This post does not claim a local DeepSeek-V4-Flash-0731 benchmark because the model weights are not present on the Thor machines and the official serving path targets multi-GPU systems.