Model Details
DeepSeek-V4 Flash Max
Quick answer: DeepSeek-V4 Flash Max is DeepSeek's February 2026 cost-efficient frontier model, scoring 91.6% on LiveCodeBench, 86.2% on MMLU-Pro, 45.1% on HLE, and 79% on SWE-Bench Verified. It offers near-V4 Pro Max performance at lower cost.
At a Glance
Where DeepSeek-V4 Flash Max leads
- 91.6% LiveCodeBench — near-ceiling coding performance
- 79% SWE-Bench Verified — excellent real-world software engineering
- 86.2% MMLU-Pro — strong academic knowledge
- Lower cost than V4 Pro Max (Flash tier)
- 45.1% Humanity's Last Exam
Where it lags
- Proprietary: closed weights
- Slightly below V4 Pro Max on all benchmarks
- Text-only
Best for: Cost-optimised frontier coding at DeepSeek V4 generation; SWE-Bench-class software engineering tasks; production API with cost efficiency.
What DeepSeek-V4 Flash Max Is
DeepSeek-V4 Flash Max is the cost-efficient tier in the V4 generation — offering near-Pro Max performance at lower cost. The "Flash" naming convention (matching DeepSeek's faster/cheaper tier) indicates optimised inference with some benchmark trade-offs vs Pro Max.
Its 79% SWE-Bench Verified score is particularly strong — placing it among the top models for real-world software engineering task completion. Combined with 91.6% LiveCodeBench, it is a premier choice for coding-intensive API workloads where cost matters.
Specifications
| Field | Value |
|---|---|
| Organization | DeepSeek |
| License | Proprietary (API only) |
| Release date | February 2026 |
| Knowledge cutoff | September 2025 |
| Modality | Text only |
Pricing
Available via DeepSeek API. Flash tier pricing is lower than Pro Max — refer to DeepSeek pricing page for current rates.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| LiveCodeBench | 91.6% | Benchgen evaluation | 2026-02 |
| MMLU-Pro | 86.2% | Benchgen evaluation | 2026-02 |
| Humanity's Last Exam | 45.1% | Benchgen evaluation | 2026-02 |
| SWE-Bench Verified | 79% | Benchgen evaluation | 2026-02 |
DeepSeek-V4 Flash Max vs Alternatives
| Model | LiveCodeBench | SWE-Bench | MMLU-Pro | License |
|---|---|---|---|---|
| DeepSeek-V4 Flash Max | 91.6% | 79% | 86.2% | Proprietary |
| DeepSeek-V4 Pro Max | 93.5% | — | 87.5% | Proprietary |
| DeepSeek-V3.2 Experimental | 74.1% | — | 85% | MIT |
Flash Max vs Pro Max: -1.9pp LiveCodeBench, -1.3pp MMLU-Pro, lower cost. For budget-sensitive V4 API use: Flash Max. For maximum performance: Pro Max. For open weights: V3.2 Experimental.
Related Models & Benchmarks
- Compare: DeepSeek-V4 Pro Max · DeepSeek-V3.2 Experimental
- Benchmarks: LiveCodeBench · SWE-Bench Verified · MMLU-Pro
Frequently Asked Questions
What is DeepSeek-V4 Flash Max?
DeepSeek's February 2026 cost-efficient frontier model scoring 91.6% LiveCodeBench, 86.2% MMLU-Pro, 45.1% HLE, and 79% SWE-Bench Verified. Lower cost alternative to V4 Pro Max.Should I use V4 Flash Max or V4 Pro Max?
Flash Max for cost-optimised coding tasks — near-Pro Max performance at lower API cost. Pro Max for maximum benchmark performance regardless of cost.Specs from DeepSeek's V4 release (February 2026) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.