← All providers
DEEPSEEK / MODEL DIRECTORY

DeepSeek model matrix

Choose the model type. Compare its position. See when it arrived.

Compare DeepSeek models in our directory by task, positioning and release date.

04models
01model types
03selection tiers

Recent model landscape

02

The 10 most recent dated models · time × selection tier

Swipe to explore the timeline. Tap a model for details.

FlagshipMainstreamValue2026-04-242026-05-292026-07-032026-08-062026-09-10Time →DeepSeek V4 Pro · 2026-04-24 · Text · FlagshipDeepSeek V4 ProText · 04-24DeepSeek V4.1 Flash · 2026-09-10 · Text · ValueDeepSeek V4.1 FlashText · 09-10
DeepSeek V4.1 FlashValueModel announcement ↗
Pricing ↗

Horizontal distance follows calendar time. Vertical bands are selection tiers, not benchmark scores; labels may occupy separate tracks within a tier. Select a model to inspect its source. 2 models without a date or tier are excluded.

Input price per 1M tokens: list vs Crazyrouter
ListCrazyrouterDeepSeek V4 Pro$1.32 → $1.34DeepSeek V4 Flash$0.300 → $0.450

Complete model matrix

Text

4 models shown

Dates follow official model or family announcements. API availability may differ; unverified records are labeled.

Scroll horizontally to compare tiers.

Read the matrix. Columns show positioning, rows show release months. Blank cells mean no listed model in that combination. Unknown dates are always placed last.
Release month
FlagshipCapability first
MainstreamBalanced everyday workloads
ValueCost & throughput first
2026091 models
—
—
2026041 models
—
—
Date unverified2 models
—
—

Columns show positioning, rows show release months. Blank cells mean no listed model in that combination. Unknown dates are always placed last.

Dates prefer official model or family announcements. Where only API availability is verified, it is explicitly labeled API release, not a model debut. Directory dates are unverified. Missing dates are never inferred from names or listing dates; generic endpoints and unresolved aliases are not assigned a version launch date.

Classification reviewed

How the tiers work

These are Crazyrouter selection tiers for DeepSeek, not the vendor's official ratings. Compare within the same category and generation; newer or more expensive does not automatically mean a better fit.

Classification reviewed

Categories follow output task: text, image, video, embeddings and audio/music are compared separately; price is a cross-check, not a capability score.

Flagship: capability, complex tasks, quality or control come first; sharing a tier does not imply equal capability.

Mainstream: balances capability, latency and cost for production; speed and balanced variants are classified by use, not price alone.

Value: cost, latency or throughput come first; lower cost does not imply better quality on every task.

Coding, reasoning, speed and specialist capabilities are descriptive tags; output modalities are assessed separately, with video and audio prices read alongside specs and duration.

DeepSeek: model positioning & official evidence

Pro is flagship within V4; Flash prioritizes a smaller model, throughput and cost. Official Flash service now runs V4.1-Flash-1017; legacy V4 Flash names are not assigned that version’s launch date. Compare matching peak/off-peak rates; off-peak is half of peak.

Official reference rates below are separate from Crazyrouter prices. Price is a cross-check, not a capability score. Unverified prices are left unspecified.

  • DeepSeek V4 ProText · Flagship

    High-capability V4 variant; date marks the initial V4 family announcement.

    USD / 1M tokens: input 1.32, output 3.96. Peak; off-peak $0.66 / $1.98; uncached input. Official pricing ↗

    Official source ↗
  • DeepSeek V4 Flash · aliasText · Value

    Legacy name; official service now routes to V4.1 Flash, prioritizing cost and throughput.

    USD / 1M tokens: input 0.3, output 1.2. Current Flash peak; off-peak $0.15 / $0.60; uncached input. Official pricing ↗

    Official source ↗
  • DeepSeek V4.1 FlashText · Value

    The version behind the current official Flash service; small, high-throughput and low-cost.

    Official source ↗
  • DeepSeek V4 Flash Vision · aliasText · Value

    Legacy vision alias now routed to V4.1 Flash; not image generation.

    USD / 1M tokens: input 0.3, output 1.2. Current Flash peak; uncached input. Official pricing ↗

    Official source ↗

Guides and comparisons

Podcast

AI Dev Tools: The Crazyrouter Podcast

Short episodes for developers on AI APIs, gateways, cost control and coding tools.

Latest episodes

  • EP196 · Oct 29, 2026 · 6:53

    AI API Security Posture — Turn Controls Into Continuous Evidence

    A practical guide to AI API security posture: inventory exposure, verify controls continuously, protect keys and data, constrain tools, and turn security assumptions into evidence that operators can act on.

  • EP195 · Oct 28, 2026 · 7:21

    AI API Policy Drift — Detect When Production Rules Stop Matching Intent

    A practical guide to policy drift in AI APIs: version routing and safety rules, detect stale enforcement, compare intended and observed behavior, and restore control before small gaps become incidents.

Episodes with written show notes →

Questions & answers

Is DeepSeek cheaper here than on DeepSeek's site?
For V4 Pro yes ($0.67 / $2.00 vs $1.32 / $3.96); V4 Flash is at list. Prices are live on each card.
Does reasoning output work?
Yes, `reasoning_content` is returned as DeepSeek documents it.