On 31 July 2026, DeepSeek announced DeepSeek V4 Flash 0731. The 0731 checkpoint superseded V4 Flash DSpark preview and produced large agent-score gains without increasing the active model size.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 31 July 2026 |
| Availability or weight release | 31 July 2026 as MIT-licensed official weights |
| Release type | official open-weight agent model replacing a preview checkpoint |
| Access | MIT-licensed weights with official vLLM and SGLang deployment recipes |
| Architecture | a 284B-total, 13B-active V4 MoE with DSpark speculative decoding |
| Maximum stated context | one million tokens, with up to 384K output recommended for high and max reasoning |
What changed
DeepSeek’s official Flash release demonstrates why preview and final model IDs cannot be collapsed. Terminal-Bench rose from 61.8 to 82.7 in the published comparison, CyberGym from 38.7 to 76.7 and DeepSWE from 7.3 to 54.4. The architecture remained the same, so the gains largely reflect post-training and agent behaviour rather than a new parameter scale.
The practical comparison is therefore not simply whether DeepSeek V4 Flash 0731 has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| Terminal-Bench 2.1 | 82.7 versus 61.8 preview | Official-to-preview gain under DeepSeek’s setup |
| CyberGym | 76.7 versus 38.7 preview | Cyber-reasoning improvement |
| DeepSWE | 54.4 versus 7.3 preview | Long-horizon repository-work gain |
These are release-time results, not independently reproduced guarantees. DeepSeek used its forthcoming minimal harness at max reasoning for code-agent tasks, while DSBench rows are internal and should not be treated as public reproduction. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
DeepSeek V4 Flash 0731 is described as a 284B-total, 13B-active V4 MoE with DSpark speculative decoding with one million tokens, with up to 384K output recommended for high and max reasoning of stated context. Its access position at verification time is MIT-licensed weights with official vLLM and SGLang deployment recipes. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For DeepSeek V4 Flash 0731, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
V4 Flash 0731 is the DeepSeek checkpoint teams should evaluate for current agent work. Registries should deprecate the preview explicitly while retaining it for historical comparison.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: DeepSeek official model card via Hugging Face. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
