# DeepSeek-V4 Flash Vision Exp ![Organization](https://img.shields.io/badge/Org-DeepSeek-555) ![Modality](https://img.shields.io/badge/Modality-Multimodal-8957e5) ![License](https://img.shields.io/badge/License-MIT-d29922) ![Released](https://img.shields.io/badge/Released-Sep%202026-555) > **Quick answer:** DeepSeek-V4-Flash-Vision-Exp is DeepSeek's first experimental multimodal model in the V4 family, built on the DeepSeek-V4-Flash architecture with added visual encoder/aligner modules and continued training. It improves multimodal agent capability substantially over the text-only DeepSeek-V4-Flash-0731 baseline while holding text-only agent performance roughly steady. Released under the MIT license, September 2026. ## At a Glance **Where it leads** - Clear multimodal-agent gains over its own text-only predecessor: ApexBench Pass@1 jumps to 36.5% (from 26.2% when the predecessor ignores image input), Agents' Last Exam to 27.3% (from 25.2%) - Text-only agent capability held essentially flat vs. DeepSeek-V4-Flash-0731 (e.g. Terminal Bench 2.1: 83.9% vs 82.7%), so vision was added without a text-capability tax - ZeroBench (Pass@5) at 35.0% edges out Claude Opus 4.8's 34.0% — notable for an experimental release **Where it lags** - Still trails Claude Opus 4.8 on most text-agent benchmarks (NL2Repo 57.7% vs 69.7%, DSBench-Hard 63.6% vs 71.7%) - Explicitly labeled "Exp" (experimental) by DeepSeek — not positioned as a production-grade flagship **Best for:** teams evaluating early multimodal agent capability from DeepSeek's V4 line before a non-experimental release ships. ## What DeepSeek-V4 Flash Vision Exp Is DeepSeek-V4-Flash-Vision-Exp is DeepSeek's first attempt at bringing visual understanding into its V4-Flash architecture. Rather than a from-scratch multimodal model, it starts from the existing DeepSeek-V4-Flash-0731 checkpoint, adds a vision encoder and aligner, and continues training to unlock image understanding — a lower-cost path to multimodality that trades off against the ceiling a from-scratch multimodal design might reach. The released repository ships a minimal reference PyTorch implementation covering the vision encoder/aligner, "DFlash" attention, MoE routing, Hyper-Connections, and the "DSpark" speculative-decoding forward path — DeepSeek publishes this alongside the weights specifically to let others reproduce inference rather than relying solely on hosted API access. ## Specifications | Field | Value | |---|---| | Organization | DeepSeek | | Parameters | Undisclosed (built on DeepSeek-V4-Flash architecture) | | Architecture | DeepSeek-V4-Flash + vision encoder/aligner, DFlash attention, MoE, Hyper-Connections, DSpark speculative decoding | | License | MIT | | Release date | 2026-09-01 (approximate, per HF "Updated" timestamp) | | Modality | Multimodal (text + vision) | ## Pricing Open weights — free to download and self-host under the MIT license; cost is inference/hosting only. No hosted API pricing published at release. ## Public Benchmark Scores *Scores below are self-reported by DeepSeek in the model's own HF model card, evaluated against DeepSeek-V4-Flash-0731 (its own text-only predecessor) and Claude Opus 4.8. They are not Benchgen measurements. DeepSeek evaluates using the minimal mode of its own DeepSeek Harness, max reasoning effort, temperature=1.0, top_p=0.95.* | Benchmark | Score | Source | Date | |---|---|---|---| | [Terminal-Bench 2.1](/benchmarks/laude-institute/terminalbench) | 83.9 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [NL2Repo](/benchmarks/nl2repo) | 57.7 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [CyberGym](/benchmarks/cybergym) | 75.3 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [DeepSWE](/benchmarks/deepswe) | 59.3 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [Toolathlon Verified](/benchmarks/hkust-nlp/toolathlon-verified) | 75.9 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [AutomationBench (Public)](/benchmarks/zapier/automationbench) | 25.7 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [Agents' Last Exam](/benchmarks/agents-last-exam) | 27.3 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | | [ZeroBench (Pass@5)](/benchmarks/zerobench) | 35.0 | [DeepSeek model card](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp) | 2026-09 | **Not mapped (excluded, no matching Benchgen benchmark entity):** DSBench-Hard, ApexBench (Pass@1 multimodal variant), Chartography. These are novel/uncommon enough that no local benchmark config exists yet; excluded rather than force-mapped to a similarly-named entity. Given this is DeepSeek's first *experimental* multimodal release (not a flagship), new benchmark configs were not stood up for these three — revisit if a non-experimental successor or other labs start reporting the same benchmarks. ## DeepSeek-V4 Flash Vision Exp vs Alternatives | Model | Terminal-Bench 2.1 | DeepSWE | Toolathlon Verified | |---|---|---|---| | **DeepSeek-V4 Flash Vision Exp** | 83.9% | 59.3% | 75.9% | | [DeepSeek-V4 Flash Max](/models/deepseek/deepseek-v4-flash-max) | — | — | — | | [Claude Opus 4.8](/models/anthropic/claude-opus-4-8) | 85.0% | 58.0% | 76.2% | *Rival scores are as reported in DeepSeek's own comparison table (same source as this model's scores above), not independently re-verified by Benchgen.* DeepSeek-V4-Flash-Vision-Exp lands close to Claude Opus 4.8 on several text-agent benchmarks despite being an experimental add-on to an existing text model — the interesting signal here isn't raw capability but that DeepSeek achieved it without a meaningful text-capability regression from adding vision. ## How DeepSeek-V4 Flash Vision Exp Performs on Real Agent Tasks The clearest agent signal is the multimodal-vs-ignoring-images comparison: DeepSeek-V4-Flash-0731 (the text-only predecessor) is evaluated on the same multimodal benchmarks by simply ignoring image inputs, and DeepSeek-V4-Flash-Vision-Exp beats it meaningfully (ApexBench 36.5% vs 26.2%, Agents' Last Exam 27.3% vs 25.2%). That's a genuine capability delta, not just a benchmark artifact — but builders evaluating this for production agent work should weigh the "Exp" label: DeepSeek has not positioned this as a finished, stable release, and the underlying parameter count and training details remain largely undisclosed. ## Related Models & Benchmarks - Compare: [DeepSeek-V4 Flash Max](/models/deepseek/deepseek-v4-flash-max) · [Claude Opus 4.8](/models/anthropic/claude-opus-4-8) - Benchmarks: [DeepSWE](/benchmarks/deepswe) · [Toolathlon Verified](/benchmarks/hkust-nlp/toolathlon-verified) · [AutomationBench](/benchmarks/zapier/automationbench) ## Frequently Asked Questions

What is DeepSeek-V4 Flash Vision Exp?

It's DeepSeek's first experimental multimodal model in the V4 family — the existing DeepSeek-V4-Flash text architecture with added vision encoder/aligner modules and continued training.

How much does DeepSeek-V4 Flash Vision Exp cost?

It's open weights under the MIT license — free to download and self-host. No hosted API pricing has been published.

Is DeepSeek-V4 Flash Vision Exp open source?

Yes, released under the MIT license, with a reference PyTorch inference implementation published alongside the weights.

Does adding vision hurt DeepSeek-V4 Flash's text performance?

No — DeepSeek reports text-only agent benchmarks holding roughly flat vs. the text-only DeepSeek-V4-Flash-0731 predecessor (e.g. 83.9% vs 82.7% on Terminal-Bench 2.1).

--- *Specs and scores sourced from DeepSeek's official model card. Third-party comparison scores attributed inline. Last updated 2026-09-08.*

DeepSeek-V4 Flash Vision Exp — Training jobs

Training jobs