DeepSeek V4.1 Flash (Alibaba)

DeepSeek V4.1 Flash is positioned as a cost- and speed-optimized sibling within DeepSeek's V4 lineup, designed as a practical substitute for the larger V4-Pro. It uses a Mixture of Experts architecture that activates only a slice of its network per token, with 552 billion parameters in the backbone but only 8 billion active during input processing and 16 billion during output generation. Compared with the earlier V4-Flash, which had 284 billion parameters and 13 billion active, V4.1 Flash broadens the backbone substantially while reshaping the active footprint to favor lighter inference.

The model was reported to outpace DeepSeek's own V4-Pro on several agentic and coding benchmarks, and on some tests it exceeded frontier systems such as GPT-5.6 Sol, signaling strength in tool-driven workflows and code generation rather than raw scale. A reduced key-value cache complements the lower active-parameter count to keep latency and serving cost in check, and DeepSeek announced it would consolidate its V4-Pro API service onto V4.1 Flash, reinforcing its role as the default efficient coding and reasoning option in the family.

Eden AIqwen/deepseek-v4.1-flashdeepseek-flash

Open provider page Provider docs

Quick Info

Provider: Eden AI
Model key: qwen/deepseek-v4.1-flash
Release date: Sep 10, 2026
Last updated: Sep 10, 2026
Knowledge cutoff: 2025-05
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.15
Output token cost: $0.60

Limits

Output tokens: 384,000 tokens
Context window: 1,000,000 tokens

Transparent token rates

Compare DeepSeek V4.1 Flash (Alibaba) pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about DeepSeek V4.1 Flash (Alibaba)

Eden AI

CoverageBenchmark

DeepSeek released DeepSeek V4.1 Flash on September 10, 2026, and the supplied excerpt from aiintoai.vercel.app describes its architecture in detail: an asymmetric causal encoder-decoder Mixture-of-Experts design with 552B total parameters, activating only 8B parameters during input prefill and 16B parameters (two route The same excerpt reports coding-benchmark numbers for the V4.1 Flash variant — 67.2% on SWE-bench Verified and 91.8% on HumanEval — and frames them as competitive with Claude 3.7 Sonnet on development tasks at roughly 1/40th the API price. Local-inference guidance on the page indicates 50+ tokens per second on Apple Si

Eden AI

CoverageBenchmark

The llm-stats.com model page for DeepSeek-V4.1-Flash, as supplied in the excerpt, ranks the variant 13th on its composite LLM Stats Score with a score of 51.8 and a blended price of $0.24 per million tokens, placing it on the cost-efficiency curve between GPT OSS 120B and GPT-6 Astra. The page's capability-tier breakdo Beyond the composite and tier rankings, the page positions V4.1-Flash against sibling DeepSeek variants in the same price band, noting DeepSeek-V4-Flash-0731 at $0.066 blended with a score of 44.7 as a cheaper predecessor and DeepSeek-V4.1-Flash itself at $0.24 with a score of 51.8 as the current release. Sources for i