DeepSeekOpen Source

DeepSeek V4 Flash

Compare this model

A high-speed MoE foundation model developed by DeepSeek. It has 284B parameters (13B active), supports a 1M token context, and achieves comparable performance to the Pro version with only 27% of the FLOPs.

Parameters

2840

Context Window

1M

License

MIT

Release Date

2026-04-24

Japanese Language Capability

High-Quality JP

Multilingual model with strong Japanese language processing capabilities.

API Pricing

Input Price (per 1M tokens)

$0.14

Output Price (per 1M tokens)

$0.28

Billing Mode: standard

Strengths

    Weaknesses

      Use Cases

        Deep Analysis

        Arena Elo

        1436

        Based on 36,817 votes on BenchLM

        SWE-bench Verified

        79.0%

        In Think Max mode

        Input Price

        $0.14/1M

        Cache miss; $0.028/1M cache hit

        Output Price

        $0.28/1M

        12x cheaper than V4-Pro

        Context Window

        1M tokens

        Native; requires only ~10% FLOPs vs V3.2

        GPQA Diamond

        88.1%

        In Think Max mode

        Strengths

        • Unbeatable price-to-performance ratio at $0.28/M output tokens—roughly 90x cheaper than Claude Opus
        • Native 1M-token context window with dramatically reduced inference cost via hybrid CSA+HCA attention architecture
        • Open-weight under MIT license with self-hostable footprint (~158GB FP4+FP8) feasible on commodity GPUs

        Weaknesses

        • Significant factual recall gap—SimpleQA-Verified at 34.1% vs Pro's 57.9%—limits high-stakes knowledge tasks
        • Substantially trails Pro on complex agentic workflows (Terminal-Bench 2.0: 56.9% vs 67.9%)
        • Preview status with potentially unstable pricing and API behavior; not yet GA

        Competitor Comparison

        ModelArenaSWEGPQAPrice
        DeepSeek V4 Pro~1550 (est.)80.6%90.1%$1.74/$3.48
        GPT-5.4 xHigh~1674 (est.)N/A93.0%~$5/$30
        Claude Opus 4.6 Max~1619 (est.)80.8%91.3%~$15/$25

        DeepSeek V4 Flash is the efficiency-optimized sibling of the V4 family, a 284B-parameter Mixture-of-Experts model with only 13B active parameters per forward pass. Released as a preview on April 24, 2026, it targets the sweet spot of high-volume, cost-sensitive production workloads while retaining a native 1-million-token context window. The model's core innovation is its hybrid attention architecture combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA), which reduces single-token inference FLOPs to roughly 10% of the prior generation V3.2 at 1M context, making extremely long context economically viable.

        Flash is explicitly positioned not as a degraded Pro but as a separately trained model that achieves comparable reasoning performance on bounded tasks when given sufficient thinking budget. In Think Max mode, it scores 91.6% on LiveCodeBench and 86.2% on MMLU-Pro—within 1-3 points of the 1.6T-parameter Pro. However, the smaller expert pool means Flash drops off on two specific axes: deep factual recall (SimpleQA-Verified gap of 23.8 points) and complex multi-step agentic workflows (Terminal-Bench 2.0 gap of 11 points). DeepSeek's own documentation acknowledges this boundary explicitly.

        The pricing model is aggressively competitive: $0.14/M input and $0.28/M output tokens positions Flash roughly 90-107x cheaper than Western closed-source alternatives on output. Combined with MIT-licensed open weights and deployment support via vLLM and SGLang on both NVIDIA B200 and H100 hardware, V4 Flash represents the most cost-effective path to frontier-adjacent capability for high-throughput production systems. The recommended architecture for most teams is hybrid routing: default to Flash for 70-80% of traffic, escalate to Pro only for complex agentic and knowledge-intensive tasks.

        Analysis generated: 2026-07-17