Skip to content
ModelCap

DeepSeek V3.2

Current board #35Arena #98Open weights

DeepSeek·deepseek/deepseek-v3.2

DeepSeek V3.2 is a language model published by DeepSeek. 14 OpenRouter endpoints are published in the latest snapshot. It accepts text input and returns text. The published context limit is 164K tokens. The catalogue declares support for reasoning, tool use, structured outputs.

ModelCap Index · Current board
70.0#35
Measured · 91% support · score interval 70.0–75.4
Market Gravity
57.6
Secondary market signal
Input
$0.269
per 1M tokens
Output
$0.40
per 1M tokens
Context
164K
tokens
Providers
14
endpoints
Gateway spend
—
latest · Vercel share

Providers

Uptime measured over the last 30 minutes
ProviderInputOutputContextQuantUptime
Baidufp8Cheapest$0.207$0.311131Kfp8100.0%
StreamLakefp8$0.214$0.322128Kfp899.90%
DeepInfrafp4$0.26$0.38164Kfp498.33%
AtlasCloudfp8$0.26$0.38164Kfp8100.0%
Novitafp8$0.269$0.40164Kfp8100.0%
SiliconFlowfp8$0.259$0.42164Kfp899.92%
GMICloudfp8$0.29$0.43164Kfp8100.0%
Venice$0.33$0.48160K—81.02%
DigitalOcean$0.25$0.80164K—100.0%
Phala$1.00$1.00164K—100.0%
Alibabafp8$0.37$1.11131Kfp898.16%
Friendli$0.50$1.50164K—100.0%
Google$0.56$1.68164K—99.92%
SambaNova$3.00$4.5033K—93.01%

Hugging Face Inference Providers

Exact repository identity

Provider economics and telemetry published by the Hugging Face Router. These are operational signals, not benchmark results.

ProviderInputOutputContextTTFTThroughputFeatures
deepinfra
live
$0.26$0.38164K574 ms23.9 tok/stools
novita
live
$0.269$0.40164K1259 ms32.8 tok/stools · structured
featherless-ai
live
——————

Reasoning efforts

Best supported result can inform ModelCap

Published Max, X-High, High and other configurations for this model. These are sourced scores, not head-to-head essays.

Related community configurations

Human preference rating
ConfigurationRatingSource rankVotesSource
Thinking
deepseek-v3.2-thinking
1423
CI 1419–1426
#10441KArena
Match 100%

Reported evaluation leads (52)

Excluded from score

Structured rows reported in this exact model repository at the pinned revision below. They are useful research leads, but remain quarantined until dataset revision, harness, configuration and identity pass ModelCap’s independent admission review.

Dataset / taskMetricValueRevisionProvenance
cais/hle
hle
—40.8
Unpinned
not published
Reported
Quarantined
claw-eval/Claw-Eval
general
—42.2
Unpinned
not published
Reported
Quarantined
claw-eval/Claw-Eval
multi_turn
—31.6
Unpinned
not published
Reported
Quarantined
collinear-ai/yc-bench
medium
—125,263
Unpinned
not published
Reported
Quarantined
FutureMa/EvasionBench
evasion_bench
—66.88
Unpinned
not published
Reported
Quarantined
harborframework/terminal-bench-2.0
terminalbench_2
—39.6
Unpinned
not published
Reported
Quarantined
Idavidrein/gpqa
diamond
—82.4
Unpinned
not published
Reported
Quarantined
internlm/WildClawBench
overall
—34
Unpinned
not published
Reported
Quarantined
internlm/WildClawBench
avg_time
—549
Unpinned
not published
Reported
Quarantined
internlm/WildClawBench
avg_cost
—11.4
Unpinned
not published
Reported
Quarantined
MathArena/aime_2026
MathArena/aime_2026
—94.17
Unpinned
not published
Reported
Quarantined
MathArena/hmmt_feb_2026
MathArena/hmmt_feb_2026
—84.09
Unpinned
not published
Reported
Quarantined
mercor/apex-agents
apex-agents
—7
Unpinned
not published
Reported
Quarantined
mercor/APEX-SWE
apex-swe
—15.8
Unpinned
not published
Reported
Quarantined
open-agent-leaderboard/results
overall
—0.4158
Unpinned
not published
Reported
Quarantined
open-agent-leaderboard/results
appworld
—0.03
Unpinned
not published
Reported
Quarantined
open-agent-leaderboard/results
browsecomp_plus
—0.48
Unpinned
not published
Reported
Quarantined
open-agent-leaderboard/results
swebench
—0.64
Unpinned
not published
Reported
Quarantined
open-agent-leaderboard/results
taubench_airline
—0.28
Unpinned
not published
Reported
Quarantined
open-agent-leaderboard/results
taubench_retail
—0.65
Unpinned
not published
Reported
Quarantined

Showing 20 of 52 rows.

Other versions (4)

ModelCap Index · Current board

Methodology
70.0
Measured · 4 public benchmark observations across 4 boards
Current board
#35
Public rank basis
Measured (measured)
Index support
90.9% · strong
Index interval
70.0–75.4
Rank posterior
Not available
Top 5 / Top 10 probability
Not available / Not available
Posterior as of
Snapshot timestamp unavailable
Identity binding
Exact catalogue product · deepseek/deepseek/deepseek-v3.2/fp8 · aggregates disclosed benchmark configurations · endpoint configuration not claimed · artifact metadata deepseek-ai/DeepSeek-V3.2@a7e62ac04ecb2c0a54d736dc46601c5606cf10a6 (not the evaluation revision)
Direct benchmark observations
4
Breadth-adjusted BES point estimate
72.7
Bayesian Evidence Score · measured evidence input only

BES normalizes admitted public benchmark evidence for measured Index rows. Its legacy score and observables prior do not define the public language rank.

Observed capability
72.7
Evidence status
confirmed · 95% mass
General preference
74.6
Coding
73.3
Agents & tools
—
Reasoning
32.6
Evidence breadth
95%
Evidence coverage
91%
Market and catalogue signals · secondary, never the language Index rank
Usage (OpenRouter popularity)
41.5
Liquidity (providers × uptime)
91.1
Open reach (HF downloads)
66.9
Surface (context / tools / modalities)
71.8
Freshness
38.7
  • · OpenRouter popularity #28
  • · 14 live providers
  • · HF 30d downloads 1194806
  • · Catalogue: context 163840, reasoning, tools
  • · First seen 2025-12-01

These adoption and deployment observations remain context only. They do not change this model's ModelCap Index score or rank.

BES evidence selected · measured input only

The bounded 0–100 capability score blends each source’s competitive placement with its published achievement, then combines capability families. Evidence breadth adds a modest uncertainty adjustment; price and popularity are not part of benchmark-led rank.

Deployment readiness

Metadata index
96.0
strong · 98% metadata coverage
Not capability
Artifact reproducibility
100.0
Access & legal clarity
100.0
Deployability
98.6
Evaluation provenance
82.1

Missing: evaluationProvenance.evaluation-dataset-revisions.

A metadata completeness and deployability index—not a safety certification, quality grade, or production-readiness claim.

Market activity

Venue coverage
Market Gravity
57.6
OpenRouter weekly popularity
#28 of 298

Market Gravity uses OpenRouter’s full-catalogue popularity order. Sparse Vercel top-ten observations appear only when published and do not affect the score.

Arena preference

Status
Ranked
Source
Arena
Official rank
#98 of 385
Rating
1425.0
95% confidence interval
1421–1429
Votes
47K
Source snapshot
6 Aug 2026, 13:00 UTC
Category
overall
Identity match confidence
100%

Market Gravity breakdown

Usage55%
41.5

OpenRouter weekly popularity across the full model catalogue.

Liquidity25%
91.1

Independent providers versus the model's open or closed cohort, weighted by uptime.

Open reach15%
66.9

Hugging Face 30-day downloads for open models; neutral for closed models.

Freshness5%
38.7

Time since first public availability, on a six-month half-life.

Specification

Context window
164K tokens
Max output
66K tokens
Inputs
Text
Outputs
Text
Tokenizer
DeepSeek
Cached input tokens
$0.134 / 1M
First seen on OpenRouter
1 Dec 2025

Capabilities

  • Supported: Reasoning
  • Supported: Tool use
  • Supported: Structured output
  • Supported: Response format
  • Not supported: Moderated

Weights & access

Open weights. Downloadable weights are published under an OSI or free-culture license in the current repository metadata. This describes weight availability, not whether the full training stack qualifies as Open Source AI.

Access
Open weights
Licence
mit
Pinned revision
a7e62ac04ecb
Parameters
685.4B
Architecture
DeepseekV32ForCausalLM
Model type
deepseek_v32
Hugging Face downloads (30d)
1.2M
Downloads (all time)
21.4M
Likes
1K
Repository
deepseek-ai/DeepSeek-V3.2
Repository updated
1 Dec 2025, 11:04 UTC
Base lineage
deepseek-ai/DeepSeek-V3.2-Exp-Base