DeepSeek-V4: a million-token context that agents can actually use

DeepSeek-V4 introduces a novel architecture that slashes KV cache requirements by up to 98%, enabling AI agents to handle long-running, complex tasks with unprecedented efficiency.

Focusing on long running agentic workloads. Running a frontier open model as an agent today breaks in predictable ways. The model stops. You reprompt. The trace blows past the context budget, or the KV cache fills the GPU, or tool-call round trips degrade halfway through a long task. V4 is built to fix these known failures, and point the way for the community to follow.

This post covers three things: what the architecture does differently to make long-context inference cheap, the agent-specific post-training decisions that compound on top of it, and some takeaways from the paper that help reason about these changes.

A 1M context window is just capacity, not performance. Whether you can use it depends on the cost of every forward pass at that depth. For an agent running a long tool-use trajectory (a SWE-bench task, a multi-step browse session, a terminal session with hundreds of commands), every tool result is appended to the context, and every subsequent token pays the full attention cost against everything that came before.

Two numbers matter: single-token inference FLOPs and KV cache size. Both grow with sequence length. At 1M tokens, DeepSeek-V4-Pro requires 27% of single-token inference FLOPs compared with DeepSeek-V3.2, so it runs faster on the same hardware. It also uses 10% of the KV cache memory. V4-Flash drops these numbers even further: 10% of the FLOPs and 7% of the KV cache.

If we compare the KV cache memory against a established architecture like grouped query attention with 8 heads, stored in the usual bfloat16 format, DeepSeek v4 requires roughly 2% the cache size. This makes it much easier to deploy for very large context handling.

The efficiency gain comes from splitting attention into two mechanisms and interleaving them across layers.

Compressed Sparse Attention (CSA) compresses KV entries by 4x along the sequence dimension using softmax-gated pooling with a learned positional bias. A lightning indexer (FP4, ReLU-scored multi-head dot product) picks the top-k compressed blocks per query. It inherits the sparse-selection idea from DeepSeek Sparse Attention in V3.2, but runs it over blocks that are already 4x shorter than the original sequence. The indexer's search space shrinks with it.

Heavily Compressed Attention (HCA) compresses KV entries by 128x and drops the sparse selection. Every query attends densely to every compressed block. The compressed sequence is short enough that dense attention is cheap.

The layers alternate between CSA and HCA. Different layers carry different attention patterns, and forcing one mechanism across all of them wastes capacity. In V4-Pro's 61-layer stack, layers 0–1 are HCA, layers 2–60 alternate CSA and HCA, and the MTP block at the end runs sliding-window only.

Both paths use FP8 storage for most KV entries and BF16 only for the RoPE dimensions. The lightning indexer inside CSA runs in FP4. These storage choices compound with the compression ratios to produce the 2% KV cache figure.

Efficient long-context attention is necessary for agent workflows but not sufficient. The paper describes three post-training and infrastructure choices that target agent use cases directly.

V3.2 kept reasoning traces across tool-result rounds but discarded them whenever a new user message arrived. For an agent handling a single user turn, this was fine. For multi-turn agentic workflows, where the user sends a follow-up after the agent has already chained several tool calls, the model lost its accumulated reasoning and had to reconstruct state.

V4 preserves reasoning content across user message boundaries when the conversation contains tool calls. The model retains the complete reasoning history across all rounds, including across user turns. This allows a coherent, cumulative chain of thought over long-horizon agent tasks.

V4 introduces a |DSML| special token and an XML-based tool-call format. The XML format reduces escaping failures compared to JSON-in-string tool calls, a common failure mode when models emit nested quoted content.

The agent behavior was trained with RL against real tool environments. DeepSeek Elastic Compute (DSec) is a Rust platform that exposes four execution substrates: function calls, containers, microVMs (Firecracker), and full VMs (QEMU). A single cluster runs hundreds of thousands of concurrent sandboxes.

Specific numbers from the agent section of Table 6:

Terminal Bench 2.0: V4-Pro-Max scores 67.9, ahead of GLM-5.1 (63.5) and K2.6 (66.7).
SWE Verified: 80.6 resolved, within a point of Opus-4.6-Max (80.8).
Toolathlon: 51.8, ahead of K2.6 (50.0) and Gemini-3.1-Pro (48.8).

Four checkpoints are on the Hub: