Skip to content
ModelCap

Qwen3.5 397B A17B

Current board #22Arena #73Open weights

Qwen·qwen/qwen3.5-397b-a17b

Qwen3.5 397B A17B is a language model published by Qwen. 11 OpenRouter endpoints are published in the latest snapshot. It accepts text, image, video input and returns text. The published context limit is 262K tokens. The catalogue declares support for reasoning, tool use, structured outputs.

ModelCap Index · Current board
76.9#22
Measured · 84% support · score interval 76.9–83.5
Market Gravity
41.1
Secondary market signal
Input
$0.39
per 1M tokens
Output
$2.34
per 1M tokens
Context
262K
tokens
Providers
11
endpoints
Gateway spend
—
latest · Vercel share

Providers

Uptime measured over the last 30 minutes
ProviderInputOutputContextQuantUptime
DigitalOceanCheapest$0.302$1.93131K—99.35%
Alibabafp8$0.39$2.34262Kfp8100.0%
Chutesfp8$0.45$3.00262Kfp8100.0%
DeepInfrafp8$0.45$3.00262Kfp899.02%
AtlasCloudfp8$0.55$3.50262Kfp874.17%
Phala$0.55$3.50262K—99.86%
Parasailfp8$0.50$3.60262Kfp8—
Novita$0.60$3.60262K—99.78%
StreamLake$0.60$3.60256K—99.82%
GMICloudfp8$0.60$3.60262Kfp8—
Venice$0.75$4.50128K—100.0%

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
$0.45$3.00262K244 ms66.1 tok/stools · structured
novita
live
$0.60$3.60262K1173 ms67.7 tok/stools
scaleway
live
$0.684$4.10—294 ms107.1 tok/stools · structured
ovhcloud
live
$0.71$4.25262K322 ms80.7 tok/stools · structured
featherless-ai
live
——————

Reported evaluation leads (20)

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
—28.7
Unpinned
not published
Reported
Quarantined
cais/hle
hle
—48.3
Unpinned
not published
Reported
Quarantined
claw-eval/Claw-Eval
general
—57.8
Unpinned
not published
Reported
Quarantined
claw-eval/Claw-Eval
multimodal
—20.8
Unpinned
not published
Reported
Quarantined
claw-eval/Claw-Eval
multi_turn
—52.6
Unpinned
not published
Reported
Quarantined
collinear-ai/yc-bench
medium
—90,787
Unpinned
not published
Reported
Quarantined
harborframework/terminal-bench-2.0
terminalbench_2
—52.5
Unpinned
not published
Reported
Quarantined
Idavidrein/gpqa
diamond
—88.4
Unpinned
not published
Reported
Quarantined
internlm/WildClawBench
overall
—34.5
Unpinned
not published
Reported
Quarantined
internlm/WildClawBench
avg_time
—459
Unpinned
not published
Reported
Quarantined
internlm/WildClawBench
avg_cost
—22.2
Unpinned
not published
Reported
Quarantined
InternScience/ResearchClawBench
overall
—14.23
Unpinned
not published
Reported
Quarantined
likaixin/ScreenSpot-Pro
overall
—65.6
Unpinned
not published
Reported
Quarantined
MathArena/aime_2026
MathArena/aime_2026
—93.33
Unpinned
not published
Reported
Quarantined
MathArena/hmmt_feb_2026
MathArena/hmmt_feb_2026
—87.88
Unpinned
not published
Reported
Quarantined
mercor/apex-agents
apex-agents
—13.6
Unpinned
not published
Reported
Quarantined
MME-Benchmarks/Video-MME-v2
video-mme-v2
—55.9
Unpinned
not published
Reported
Quarantined
MME-Benchmarks/Video-MME-v2
video-mme-v2
—48.9
Unpinned
not published
Reported
Quarantined
SWE-bench/SWE-bench_Verified
swe_bench_%_resolved
—76.4
Unpinned
not published
Reported
Quarantined
TIGER-Lab/MMLU-Pro
mmlu_pro
—87.8
Unpinned
not published
Reported
Quarantined

ModelCap Index · Current board

Methodology
76.9
Measured · 2 public benchmark observations across 2 boards
Current board
#22
Public rank basis
Measured (measured)
Index support
84.2% · strong
Index interval
76.9–83.5
Rank posterior
Not available
Top 5 / Top 10 probability
Not available / Not available
Posterior as of
Snapshot timestamp unavailable
Identity binding
Exact catalogue product · qwen/qwen/qwen3.5-397b-a17b · aggregates disclosed benchmark configurations · endpoint configuration not claimed · artifact metadata Qwen/Qwen3.5-397B-A17B@8472618112abcbd45acbcdc58436aff4233c23f7 (not the evaluation revision)
Direct benchmark observations
2
Breadth-adjusted BES point estimate
80.2
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
80.2
Evidence status
confirmed · 83% mass
General preference
80.0
Coding
81.1
Agents & tools
—
Reasoning
—
Evidence breadth
92%
Evidence coverage
84%
Market and catalogue signals · secondary, never the language Index rank
Usage (OpenRouter popularity)
16.7
Liquidity (providers × uptime)
88.6
Open reach (HF downloads)
47.8
Surface (context / tools / modalities)
90.3
Freshness
51.8
  • · OpenRouter popularity #115
  • · 11 live providers
  • · HF 30d downloads 389957
  • · Catalogue: context 262144, reasoning, tools, modalities Text/Image/Video
  • · First seen 2026-02-16

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
85.3
strong · 94% metadata coverage
Not capability
Artifact reproducibility
85.0
Access & legal clarity
100.0
Deployability
96.6
Evaluation provenance
54.2

Missing: artifactReproducibility.base-lineage, 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
41.1
OpenRouter weekly popularity
#115 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
#73 of 385
Rating
1442.2
95% confidence interval
1439–1446
Votes
64K
Source snapshot
6 Aug 2026, 13:00 UTC
Category
overall
Identity match confidence
100%

Market Gravity breakdown

Usage55%
16.7

OpenRouter weekly popularity across the full model catalogue.

Liquidity25%
88.6

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

Open reach15%
47.8

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

Freshness5%
51.8

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

Specification

Context window
262K tokens
Max output
66K tokens
Inputs
Text, Image, Video
Outputs
Text
Tokenizer
Qwen3
Cached input tokens
Not offered
First seen on OpenRouter
16 Feb 2026

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
apache-2.0
Pinned revision
8472618112ab
Parameters
403.4B
Architecture
Qwen3_5MoeForConditionalGeneration
Model type
qwen3_5_moe
Hugging Face downloads (30d)
390K
Downloads (all time)
5.0M
Likes
2K
Repository
Qwen/Qwen3.5-397B-A17B
Repository updated
24 Apr 2026, 05:51 UTC