deepseek-v4-pro

Aikido's security-focused benchmark study (published August 21, 2026) burned roughly 11.7 billion tokens testing ten contenders across three attempts on 32 fresh, off-the-shelf vulnerabilities. The lineup added GLM-5.3, DeepSeek V4 Pro 0813, DeepSeek V4 Flash 0731, Qwen3.8-Max, Kimi K3, and Grok 4.6 to its evaluation, The study concludes that open-source models now outperform the public closed frontier on pooled vulnerability recall, with DeepSeek V4 Pro topping every public closed model tested and Qwen, Kimi, and GLM-5.3 offering strong consistency. It also flags a trade-off: while coverage rose, the cheaper open models produced th

Quartz's article reports the official general-availability launch of DeepSeek-V4-Pro-0813 on Thursday, August 13, 2026, following an April preview period. It states that the GA checkpoint focuses on agent capabilities — tool use, code execution, and multi-step workflows — and lists DeepSeek's own benchmark numbers: Ter The piece additionally documents a pricing restructure effective 16:00 UTC on August 16, 2026: V4-Pro output tokens were set to rise to $3.96 per million at peak hours from a prior flat rate of $0.87 per million, alongside a new peak/off-peak billing split. These are model-level capability and pricing facts drawn from

SCMP offers third-party characterization of the DeepSeek-V4-Pro-0813 update, reporting that developer reception of the general benchmark scores was muted while researchers flagged standout performance in cybersecurity-adjacent workloads. The article notes that DeepSeek's own announcement page briefly carried language d The SCMP piece is useful as a contrarian signal alongside DeepSeek's self-reported agent benchmark numbers and QZ's neutral launch coverage. It does not introduce new technical specifications or API details, but it does temper the official benchmark narrative with independent developer sentiment, which is relevant cont

The release-tracker page documents DeepSeek-V4-Pro-0813, released on August 13, 2026, as a model-level update from DeepSeek itself. It enumerates the benchmark suites used to evaluate the model — BullshitBench v2, SWE-Bench Verified, Terminal-Bench 2.1, Toolathlon-Verified, BrowseComp, CyberGym, and two additional test The page is centered on the model itself rather than on any serving provider, which is appropriate for a model-focused subject. It also lists DeepSeek as the operator running the model and answering requests, with context-window and provider-availability fields that go beyond pure pricing. There is no Azure-specific ho

Eigent published a developer-focused workflow demonstrating how to automate monthly engineering reports using deepseek-v4-pro:cloud through its agentic platform, eliminating the manual GitHub-PR-to-Slack reporting routine. The setup requires generating an API key from the ollama.ai dashboard, then in Eigent's model set The tutorial then walks through building a dedicated Eigent worker equipped with the Slack tool and authenticating it to a workspace so the worker can post on the team's behalf. Once configured, a single prompt drives the worker to pull the previous month's merged pull requests from GitHub, compile a structured Word do

Ollama Cloud added DeepSeek-V4-Pro to its cloud model library as a frontier-tier model accessed via the deepseek-v4-pro:cloud tag. According to the technical write-up, the model is a 1.6 trillion parameter Mixture-of-Experts checkpoint with 49 billion active parameters per token and a 1 million token context window — a The release ships with pre-wired integrations into major agentic coding harnesses — Claude Code, Codex, OpenCode, OpenClaw, and Hermes Agent — so developers can route those tools to DeepSeek-V4-Pro through Ollama Cloud without extra glue. Usage patterns documented include the `ollama run deepseek-v4-pro:cloud` CLI comm

DeepSeek-chat and deepseek-reasoner retire July 24, 2026. Step-by-step migration to deepseek-v4-pro and deepseek-v4-flash with code diffs.

The DeepSeek API change log dated September 10, 2026 announces DeepSeek-V4.1-Flash as the smallest model in a new architecture family with native multimodal visual understanding, and explicitly states that V4.1 Flash now outperforms DeepSeek V4 Pro across performance, cost, speed, and total time. It documents a concret Critically for the DeepSeek-V4-Pro subject, the changelog announces an orderly retirement of V4 Pro: after 12:00 Beijing Time on September 14, 2026, and until the future release of V4.1 Pro, all requests to deepseek-v4-pro will be routed to V4.1 Flash and billed at V4.1 Flash pricing. This is a model-level behavioral c