DeepSeek releases V4 Flash and V4 Pro with 1M token context
DeepSeek unveiled V4 Flash and V4 Pro on April 24, featuring 1 million token context windows and a hybrid attention architecture. The models target agentic tasks and coding benchmarks at significantly lower cost than competitors.
DeepSeek released two variants of its V4 flagship on April 24, 2026. V4 Flash and V4 Pro both support 1 million token context windows, allowing entire codebases or long documents to be processed in a single prompt. The startup introduced a Hybrid Attention Architecture designed to improve long-context memory retention and reasoning performance.
Architecture and performance
The hybrid attention mechanism represents a technical shift in how the models handle extended sequences. DeepSeek optimized V4 for integration with popular agent frameworks, including Anthropic's Claude Code and OpenClaw. Benchmark profiles suggest the models deliver strong agentic capability at substantially lower inference cost than comparable alternatives.
Market positioning
V4 enters a crowded field of frontier models released in rapid succession. The emphasis on context window size and cost efficiency targets developers building long-horizon reasoning systems and code analysis agents. DeepSeek's approach prioritizes practical deployment constraints (token budget and latency) over raw benchmark scores.