DeepSeek models

Each model has its own URL. DeepSeek-V4.1-Flash is served as DeepSeek-V4.1-Flash-1017 on the API name deepseek-flash. Benchmark figures link to official sources where this directory publishes them. Model name migration guide and off-peak pricing guide and LLM API pricing comparison. For third-party benchmarks against Claude, Gemini, GPT and Kimi, see DeepSeek vs frontier models.

Checked on 15 Nov 2026 against DeepSeek API docs.

Model comparison (context, parameters, API pricing)

API prices apply only to models on the current DeepSeek price list. Other rows link to official model cards.

Context
1M
Params (active)
552B total · 16B active
Architecture
Asymmetric Causal MoE (6 of 384 experts)
API price (in/out per 1M)
$0.30 / $1.20
Docs
Context
1M
Params (active)
1.6T total · 49B active
Architecture
1.6T Sparse MoE (49B activated) with hybrid CSA
API price (in/out per 1M)
$1.32 / $3.96
Docs
Context
128K
Params (active)
671B total · 37B active
Architecture
DeepSeekMoE
API price (in/out per 1M)
—
Model card
Context
128K
Params (active)
671B total · 37B active
Architecture
DeepSeekMoE (256 routed
API price (in/out per 1M)
—
Model card
Context
—
Params (active)
1.5B total · dense
Architecture
SigLIP-L understanding
API price (in/out per 1M)
—
Model card

V4.1 Flash vs V4 Pro

Figures below come from this directory's model data, which mirrors the official Models & Pricing table and model cards. API model names.

Topicdeepseek-flash (V4.1-Flash-1017)deepseek-v4-pro (V4-Pro-0813)
Peak price (cache-miss in / out per 1M)$0.30 / $1.20$1.32 / $3.96
Context / max output1M / 384K max1M / 384K max
Backbone (active params)552B (8B prefill / 16B decode)1.6T (49B active)
Vision on APISupportedNot supported
Best fit (from model pages)Fast agentic coding, multimodal, lower list priceHeavier reasoning, largest Pro checkpoint

Source: Models & Pricing, V4.1-Flash-1017 model card, V4-Pro model card.

API list prices & model specs (official)

Peak and off-peak USD per 1M tokens. Off-peak is half of peak; peak hours are listed on the pricing page.

API model ID

deepseek-flash

DeepSeek-V4.1-Flash-1017

Context
1M
Max output
384K
Vision
Yes
Peak (in/out)
$0.30 / $1.20
Off-peak (in/out)
$0.15 / $0.60
Docs

API model ID

deepseek-v4-pro

DeepSeek-V4-Pro-0813

Context
1M
Max output
384K
Vision
No
Peak (in/out)
$1.32 / $3.96
Off-peak (in/out)
$0.66 / $1.98
Docs
  • DeepSeek-V4.1-Flash — Ultra-Fast Asymmetric Causal MoE with 1M Context & Engram Memory
  • DeepSeek-V4-Pro-0813 — 1.6T Sparse MoE for Autonomous Software Engineering & Heavy Reasoning
  • DeepSeek-R1 — Pioneering Open Reasoning Model Trained via Large-Scale RL & GRPO
  • DeepSeek-V3 — The Efficiency Landmark: 671B MoE with Multi-Head Latent Attention
  • Janus-Pro — Unified multimodal understanding and generation (decoupled vision encoding)

Models & architecture lineage

Compare DeepSeek-V4.1-Flash, V4-Pro (1.6T), R1, V3, and Janus-Pro. Figures link to official sources where available.

DeepSeek-V4.1-Flash

Ultra-Fast Asymmetric Causal MoE with 1M Context & Engram Memory

September 2026
sourced
Context
1M tokens
1,048,576 Tokens (1M Native)
Throughput
Not published
No single tok/s figure on the model card.
Input / output
$0.30 / $1.20
USD per 1M tokens
552B MoE backbone8B prefill / 16B decodeEngram (196B)890-byte KV cache1M contextFP4 E2M1 KVAPI: deepseek-flashVision supported

DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with a 552B-parameter backbone, a Causal Encoder-Decoder that activates 8B parameters per token during prefill and 16B during decode, a 1,048,576-token context, Engram conditional memory (196B parameters), and a global KV cache of 890 bytes per token using FP4 (E2M1) caching. On 17 October 2026 DeepSeek upgraded the served weights to DeepSeek-V4.1-Flash-1017; the API model name is still deepseek-flash and token list prices are unchanged.

Core architecture & hardware specifications

MoE Backbone
552B Total / 384 Experts
Asymmetric routing activating 6 experts (8B prefill / 16B decode)Source
Context & KV Cache
1,048,576 Tokens
890 bytes/token footprint via FP4 E2M1 quantizationSource
Engram Memory
196B Parameters
O(1) constant-time hashed associative memorySource
API price (peak)
$0.30 / $1.20
Per 1M tokens: cache-miss input / output. Cache hit $0.006. Off-peak is half.Source
Inference economics & prefix caching
Peak cache miss $0.30 / off-peak $0.15 per 1M input. Peak cache hit $0.006 / off-peak $0.003.

Peak cache miss $0.30 / off-peak $0.15 per 1M input. Peak cache hit $0.006 / off-peak $0.003.

Peak $1.20 / off-peak $0.60 per 1M output.

Price source

Sourced performance & research benchmarks

Codeforces (Rating)flagship
3,471sourced
Highest rating in the model card's published comparison table (max reasoning effort)
GPQA Diamond (Pass@1)flagship
90.9%sourced
Max reasoning effort, from the model card comparison table
Terminal-Bench 2.1 (Pass@1)flagship
90.6%sourced
Evaluated with DeepSeek Harness Minimal mode and a 1M context
HLE with tools (Pass@1)
63.9%sourced
This is the with-tools score. Text-only HLE on the same card is 36.8 (39.1 on the text-only subset).

Key innovations & capabilities

  • Asymmetric causal routing: 8B active parameters during prefill, 16B active during generation across 6 of 384 fine-grained experts.
  • Engram conditional memory: 196B parameters in constant-time O(1) hashed memory for persistent factual and codebase recall.
  • FP4 E2M1 ultra-compact KV cache: Compresses key-value state to only 890 bytes per token, making 1M token contexts computationally negligible.
  • The API supports thinking mode (on by default) and reasoning_effort values low, high, and max. The model card also describes a continuous effort setting from 1 to 100.
  • A single tokens-per-second throughput figure is not stated on the model card or the API pricing page, so none is listed here.

DeepSeek-V4-Pro-0813

1.6T Sparse MoE for Autonomous Software Engineering & Heavy Reasoning

August 2026
sourced
Context
1M tokens
1,000,000 Tokens (1M Native)
Active params
49B Active / Token
Input / output
$1.32 / $3.96
USD per 1M tokens
1.6T MoE49B active1M contextAPI: deepseek-v4-proNo vision on APIOff-peak is half price

DeepSeek-V4-Pro is the 1.6T-parameter Mixture-of-Experts model (49B activated) with a 1,000,000-token context, published on Hugging Face as DeepSeek-V4-Pro. The API serves checkpoint DeepSeek-V4-Pro-0813 as model name deepseek-v4-pro. The V4.1 model card's comparison table reports this checkpoint's benchmark scores; a separate V4.1-Pro had not launched when the 10 September 2026 API note was published.

Core architecture & hardware specifications

Total Parameters
1.6 Trillion
49B activated parameters, from the DeepSeek-V4-Pro model cardSource
Context
1M tokens
API max output 384K. Vision is not supported on this API model.Source
API model name
deepseek-v4-pro
Checkpoint DeepSeek-V4-Pro-0813Source
Inference Economics
$1.32 / $3.96 peak
Cache hit $0.044 peak. Off-peak output $1.98 per 1M tokens.Source
Inference economics & prefix caching
Peak cache miss $1.32 / off-peak $0.66 per 1M input. Peak cache hit $0.044 / off-peak $0.022.

Peak cache miss $1.32 / off-peak $0.66 per 1M input. Peak cache hit $0.044 / off-peak $0.022.

Peak $3.96 / off-peak $1.98 per 1M output.

Price source

Sourced performance & research benchmarks

Codeforces (Rating)flagship
3,348sourced
Listed for DS-V4-Pro in the V4.1-Flash model card comparison table
GPQA Diamond (Pass@1)flagship
92.4%sourced
From the same model card comparison table
Terminal-Bench 2.1 (Pass@1)flagship
87.9%sourced
DS-V4-Pro column on the V4.1-Flash model card

Key innovations & capabilities

  • 1.6 trillion total parameters and 49 billion activated parameters, from the DeepSeek-V4-Pro model card.
  • API context length 1M tokens and maximum output 384K tokens, from the Models & Pricing page.
  • Vision is not supported on deepseek-v4-pro. V4.1-Flash is the API model that lists vision.
  • Off-peak API rates are half of peak rates. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday, excluding Chinese public holidays.
  • SWE-bench Verified is not listed in the V4.1-Flash model card comparison table used for sourced figures on this page.

DeepSeek-R1

Pioneering Open Reasoning Model Trained via Large-Scale RL & GRPO

January 2025
sourced
Context
128K Tokens
Parameters
671B Total
API list price
Not on current price list
671B MoE37B ActiveGRPO Without CriticNative CoTMATH-500: 97.3%AIME: 79.8%Open Weights

DeepSeek-V3

The Efficiency Landmark: 671B MoE with Multi-Head Latent Attention

December 2024
sourced
Context
128K Tokens
Parameters
671B Total
API list price
Not on current price list
671B MoE37B ActiveMLA AttentionFP8 Mixed PrecisionAux-Loss-Free128K ContextOpen Weights

Janus-Pro

Unified multimodal understanding and generation (decoupled vision encoding)

February 2025
Context
Not stated on the Janus-Pro-7B model card
Parameters
1.5B and 7B variants
API list price
Not on current price list
Decoupled multimodalSigLIP-L (384px)Discrete VQ generationOpen weights

Comprehensive frontier benchmark matrix

Sourced figures only
V4.1-Flash-1017
AIME
—
SWE-bench
—
Codeforces
3,471
Term.-bench
90.6%
Key
GPQA Diamond 90.9% · HLE with tools 63.9%
V4-Pro-0813
AIME
Not in the sourced comparison table
SWE-bench
—
Codeforces
3,348
Term.-bench
87.9%
Key
1.6T / 49B active · GPQA Diamond 92.4%
R1
AIME
79.8%
SWE-bench
—
Codeforces
2,029
Term.-bench
—
Key
MATH-500 97.3% (R1 README)
V3
AIME
—
SWE-bench
—
Codeforces
—
Term.-bench
—
Key
MMLU 88.5 · HumanEval-Mul 82.6 · GSM8K 89.3 (V3 README)
Janus-Pro
AIME
—
SWE-bench
—
Codeforces
—
Term.-bench
—
Key
See Hugging Face model card (architecture only)

Cells marked — are not published on the official model card.