DeepSeek V3.2

DeepSeekKey deepseek-v3.2Also reported as deepseek/deepseek-v3.2, deepseek-ai/deepseek-v3.2, pro/deepseek-ai/deepseek-v3.2, deepseek-ai/DeepSeek-V3.2

Rank
#819
by active users
Active users
1
No change from 1 active users in the previous 30 days. vs previous 30 days
Adoption
0.1%
of 958 accounts
Tokens
67.1K
cache included
Est. cost
<$1
at public list prices
First seen
Dec 1, 2025
in any submission

Oct 14 – Nov 12, 2026 (UTC)958 accounts100% of tokens from an identified modelData through Nov 12, 2026, 02:53 UTCHow we count

Daily active users

Distinct accounts that used this model on each UTC day.

0 active users on Nov 12, 2026

Daily active users for DeepSeek V3.2, Oct 14, 2026 to Nov 12, 2026. Peak 1. Use the left and right arrow keys to read each day.110
Show daily values
Daily active users and tokens for DeepSeek V3.2
Date (UTC)Active usersTokens
Oct 14, 202600
Oct 15, 2026167.1K
Oct 16, 202600
Oct 17, 202600
Oct 18, 202600
Oct 19, 202600
Oct 20, 202600
Oct 21, 202600
Oct 22, 202600
Oct 23, 202600
Oct 24, 202600
Oct 25, 202600
Oct 26, 202600
Oct 27, 202600
Oct 28, 202600
Oct 29, 202600
Oct 30, 202600
Oct 31, 202600
Nov 1, 202600
Nov 2, 202600
Nov 3, 202600
Nov 4, 202600
Nov 5, 202600
Nov 6, 202600
Nov 7, 202600
Nov 8, 202600
Nov 9, 202600
Nov 10, 202600
Nov 11, 202600
Nov 12, 202600

Token mix

How the 67.1K tokens split. Cache reads are usually most of an agent's tokens and are priced far below input.

  • Input50%
  • Output0.3%
  • Cache read50%
  • Cache write0%
  • Reasoning0%

Agents that carry DeepSeek V3.2

Each agent's share of this model's tokens, and how many accounts used the model through it.

  1. OpenCode100% of this model's tokens1active users

Benchmarks

The best result for DeepSeek V3.2 in each benchmark Tokscale tracks. Each benchmark has its own unit, run conditions and source, so compare a score only with other models on the same benchmark. Rank is among the models Tokscale matched in that benchmark.

All benchmarks
  • WebDev ArenaWeb development · LMArena
    1,362
    Rank #86 of 110 matched modelsNo conditions reportedPublished Oct 19, 2026Source: huggingface.co (opens in a new tab)
    Show 1 more result with other run conditionsHide the other results
    • Listed as DeepSeek-V3.2
      1,326
      No conditions reportedPublished Oct 19, 2026Source: huggingface.co (opens in a new tab)

Not evaluated: Epoch Capabilities Index, GPQA Diamond, Humanity's Last Exam, FrontierMath Tiers 1–3 (v2), SimpleQA Verified, Terminal-Bench 4.0, CursorBench, DeepSWE, FrontierSWE, Agent Arena. A missing result is not a score of zero.

Pricing

List prices from models.dev in US dollars per 1M tokens, retrieved Oct 21, 2026. Amazon Bedrock is the reference price for the estimates below. It is the first provider listed; models.dev lists no price from the lab's own API.

Price at real agent usageEstimate at list prices

What 1M tokens cost at the reference list price when they split into input, output and cache the way Tokscale users' agents used them. It is not what anyone paid.

At this model's mix
$0.62
Est. per 1M tokens
50% cache reads
At the community mix
$0.62
Est. per 1M tokens
96% cache reads
3:1 input/output blend
$0.93
Per 1M tokens at list prices
No cache

Cache reads were 50% of DeepSeek V3.2 tokens in the last 30 days and 96% of all tokens. The 3:1 blend, (3 × input + output) ÷ 4, is the usual comparison and ignores cache.

Assumptions

  • Amazon Bedrock list prices per 1M tokens: input $0.62, output $1.85.
  • This model's mix is its own input, output, cache and reasoning tokens in the last 30 days.
  • The community mix is the tokens of every identified model in the last 30 days.
  • This is an estimate at list prices, not what anyone paid.
  • Reasoning tokens are priced as output.
  • Long-context tiers, batch pricing and discounts are not applied.
  • Token mixes come from public Tokscale submissions. Hidden accounts are excluded.
  • No cache-read price is listed, so cache reads are priced as input.
  • No cache-write price is listed, so cache writes are priced as input.

List prices by provider

  • Amazon BedrockReferencedeepseek.v3.2
    $0.62 input per 1M tokens$1.85 output per 1M tokensCache read not listedCache write not listed164K tokens of context
  • OpenRouterdeepseek/deepseek-v3.2
    $0.28 input per 1M tokens$0.42 output per 1M tokens$0.03 cache read per 1M tokensCache write not listed164K tokens of context
  • Vercel AI Gatewaydeepseek/deepseek-v3.2
    $0.62 input per 1M tokens$1.85 output per 1M tokensCache read not listedCache write not listed128K tokens of context

Specs

From models.dev (MIT), retrieved Oct 21, 2026, as the lab or the first listed provider reports them. A dash means models.dev does not list the value.

Context window
164K tokens
Max output
66K tokens
Input
Text
Output
Text
Knowledge cutoff
Jul 2024
Release date
Dec 1, 2025
Open weights
Yes
Reasoning
Yes
Tool calling
Yes

Top public profiles

Public profiles that used DeepSeek V3.2 most in this period, by tokens. Hidden accounts are never listed.

  1. 1BitmanAlan@BitmanAlan67.1Ktokens<$1est. cost1active days<$1 est. cost · 1 active days

More from DeepSeek

Other DeepSeek models in the same period, by active users.

  1. DeepSeek V4.1 Flash23% adoption225active users
  2. DeepSeek V4.1 Flash Free15% adoption141active users
  3. DeepSeek Flash13% adoption121active users
  4. DeepSeek V4.1 Flash 10177.9% adoption76active users
  5. DeepSeek V4 Flash5.3% adoption51active users
  6. DeepSeek V4 Pro5.3% adoption51active users
All models

Sources and attribution

Each score keeps its own source, unit and run conditions. Tokscale does not combine benchmarks into one score, and a missing score means the model was not evaluated, not that it scored zero.

  • LMArenaWebDev Arena and Agent Arena leaderboards.CC BY 4.0 · Retrieved Oct 21, 2026 · Data through Oct 19, 2026LMArena leaderboard dataset (lmarena-ai/leaderboard-dataset), Hugging Face.
  • models.devSpecs and list prices per provider.MIT · Retrieved Oct 21, 2026models.dev, an open catalog of AI models (github.com/anomalyco/models.dev).
  • TokscaleActive users, tokens and token mix from public Tokscale submissions. Hidden accounts are excluded.How we count