{"slug": "deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output", "title": "DeepSeek V4 Pro: 80.6% SWE-Bench at $0.87/M Output", "summary": "DeepSeek V4 Pro, a Mixture-of-Experts model with 1.6 trillion total parameters, scores 80.6% on SWE-bench Verified, tying Gemini 3.1 Pro and achieving the highest score for any open-weight model. Priced at $0.87 per million output tokens, it is 28.7 times cheaper than Claude Opus 4.8, but a NIST CAISI evaluation found it trails US frontier models on complex agentic tasks and responded to 94% of malicious jailbreak requests.", "body_md": "DeepSeek V4 Pro scores 80.6% on SWE-bench Verified — tied with Gemini 3.1 Pro and the highest score any open-weight model has posted. It ships a 1-million-token context window, runs under the MIT license, and costs $0.87 per million output tokens. That is 28.7 times cheaper than Claude Opus 4.8. It has been available since April 24, 2026, and most engineering teams have not evaluated it yet.\n\n## What V4 Pro Actually Is\n\nV4 Pro is a Mixture-of-Experts model with 1.6 trillion total parameters, activating 49 billion per token. DeepSeek built it around two architectural choices that matter for agentic workloads: a hybrid Compressed Sparse Attention mechanism that cuts inference FLOPs to 27% of its predecessor at 1-million-token contexts, and Multi-head Latent Attention that reduces KV cache memory significantly. The result is a model that does not balloon in cost as your agent context grows — a real constraint with every other frontier-class model.\n\nThe weight file is 862 GB in FP8. Open weights are on [Hugging Face](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro). The full architecture and training details are in the [DeepSeek-V4 technical paper](https://arxiv.org/html/2606.19348v1).\n\n## The Benchmark That Matters\n\nSWE-bench Verified resolves real GitHub issues from real repositories. It is the benchmark that correlates best with what coding agents actually do in production. At 80.6%, V4 Pro ties Gemini 3.1 Pro — a closed, frontier-priced model. On DeepSeek’s internal R&D coding benchmark, V4 Pro posts a 67% pass rate versus Claude Sonnet 4.5’s 47%.\n\nThat is a significant gap on the metric your CI pipeline cares about, not a synthetic puzzle benchmark.\n\n## The Price Gap Is a Category Change\n\nHere is the comparison worth running before your next model contract renewal:\n\n| Model | Output $/M tokens | Multiple vs V4 Pro |\n|---|---|---|\n| DeepSeek V4 Flash | $0.28 | 0.3x |\n| DeepSeek V4 Pro | $0.87 | 1x |\n| Claude Sonnet 4.5 | ~$3.00 | 3.4x |\n| Claude Opus 4.8 | ~$24.98 | 28.7x |\n| GPT-5.5 | ~$30.02 | 34.5x |\n\nAt 10 million output tokens per month — a mid-sized coding agent running automated PR reviews or test generation — V4 Pro costs roughly $8,700. The same workload on Claude Opus 4.8 runs approximately $249,800. That is not a line item to optimize; it changes the economic case for agent deployment entirely.\n\n## Two Catches Worth Knowing\n\nFirst: the [NIST CAISI evaluation from May 2026](https://www.nist.gov/news-events/news/2026/05/caisi-evaluation-deepseek-v4-pro) found that V4 Pro trails the US frontier by roughly eight months on complex agentic tasks (44% vs GPT-5.5’s 78% on the CAISI agentic benchmark). More critically for enterprise teams: it responded to 94% of malicious jailbreak requests in the evaluation, compared to 8% for US reference frontier models. If you operate in a regulated industry or handle sensitive data, that number blocks a production deployment without guardrails.\n\nSecond: if you self-host, the chat template issue will bite you. V4 Pro has no standard HuggingFace Jinja chat template. Naive integrations that assume one will produce silently malformed prompts — the kind that look fine until your agent starts misrouting tool calls. Use the encoding scripts that ship in the [HuggingFace repo](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro), not generic templates. The [vLLM integration guide](https://vllm.ai/blog/deepseek-v4) covers this specifically.\n\n## Who Should Evaluate It Now\n\nIf you are running coding agents at scale — automated code review, test generation, documentation, CI repair — and your current model costs are a meaningful budget line, V4 Pro is worth evaluating this week. The API is live, the SWE-bench numbers are real, and the price gap is large enough that the switching cost pays for itself quickly.\n\nIn enterprise or regulated environments, evaluate it behind a guardrails layer and treat the NIST safety findings as a hard constraint, not a footnote. Self-hosting only makes economic sense past $15,000 per month in API spend — below that, the roughly $17,500-per-month infrastructure cost does not pencil out.\n\nFor most teams, the API is the right starting point. The open weights are there when you need them.", "url": "https://wpnews.pro/news/deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output", "canonical_source": "https://byteiota.com/deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output/", "published_at": "2026-07-29 22:08:30+00:00", "updated_at": "2026-07-29 22:23:13.238927+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-safety", "ai-policy"], "entities": ["DeepSeek", "DeepSeek V4 Pro", "Gemini 3.1 Pro", "Claude Opus 4.8", "NIST", "Hugging Face", "vLLM"], "alternates": {"html": "https://wpnews.pro/news/deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output", "markdown": "https://wpnews.pro/news/deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output.md", "text": "https://wpnews.pro/news/deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output.txt", "jsonld": "https://wpnews.pro/news/deepseek-v4-pro-80-6-swe-bench-at-0-87-m-output.jsonld"}}