VENICE ATLAS · MODEL DOSSIER · QWEN 3.5 397B

Model overview

Qwen 3.5 397B

Name: Qwen 3.5 397B
Brand: venice.ai
Price: 0.7500 USD
Availability: InStock

Qwen 3.5 is Alibaba flagship reasoning model featuring a 397B parameter Mixture-of-Experts architecture with 17B active parameters. It excels at complex reasoning, coding, and general knowledge tasks.

qwen3-5-397b-a17b · venice.ai

Text · Venice catalog$0.75 per 1M tokensPrice on Venice AI

ModalityTextOn Venice AI

Primary price$0.75 per 1M tokensper 1M tokens

Context window128KMax tokens

Benchmarks38Benchmark results

Benchmark results

See the tests behind each score. They use the benchmark’s original scoring scale.

Biology#3229.88 rating Chat#129.56 rating Chemistry#2333.64 rating Communication#818.49 rating Economics#423.50 rating Finance#238.82 rating Frontend Development#2221.56 rating Healthcare#439.17 rating Instruction Following#626.91 rating Language#436.80 rating Legal#438.82 rating Long Context#1224.35 rating Math#2836.58 rating Physics#1835.00 rating Search#2119.02 rating Spatial Reasoning#2420.68 rating Structured Output#329.68 rating

Agents5 results

DeepPlanningdeep-planning · agents

34.3%

ReportedOct 2026

MCP-Markmcp-mark · agents

46.1%

ReportedOct 2026

t2-bencht2-bench · agents

86.7%

ReportedOct 2026

Toolathlontoolathlon · agents

38.3%

ReportedOct 2026

VITA-Benchvita-bench · agents

49.7%

ReportedOct 2026

Chat2 results

IFEvalifeval · chat

92.6%

ReportedOct 2026

Multi-Challengemultichallenge · chat

67.6%

ReportedOct 2026

Coding1 result

SecCodeBenchseccodebench · code

68.3%

ReportedOct 2026

General intelligence4 results

C-Evalc-eval · general

93.0%

ReportedOct 2026

Includeinclude · general

85.6%

ReportedOct 2026

MAXIFEmaxife · general

88.2%

ReportedOct 2026

NOVA-63nova-63 · general

59.1%

ReportedOct 2026

Instruction Following1 result

IFBenchifbench · instruction_following

76.5%

ReportedOct 2026

Language5 results

MMLU-Prommlu-pro · language

87.8%

ReportedOct 2026

MMLU-ProXmmlu-prox · language

84.7%

ReportedOct 2026

MMLU-Reduxmmlu-redux · language

94.9%

ReportedOct 2026

MMMLUmmmlu · language

88.5%

ReportedOct 2026

WMT24++wmt24++ · language

78.9%

ReportedOct 2026

Long Context1 result

LongBench v2longbench-v2 · long_context

63.2%

ReportedOct 2026

Math5 results

AIME 2026aime-2026 · math

91.3%

ReportedOct 2026

HMMT 2025hmmt-2025 · math

94.8%

ReportedOct 2026

HMMT25hmmt25 · math

92.7%

ReportedOct 2026

IMO-AnswerBenchimo-answerbench · math

80.9%

ReportedOct 2026

PolyMATHpolymath · math

73.3%

ReportedOct 2026

Reasoning6 results

AA-LCRaa-lcr · reasoning

68.7%

ReportedOct 2026

BrowseCompbrowsecomp · reasoning

69.0%

ReportedOct 2026

BrowseComp-zhbrowsecomp-zh · reasoning

70.3%

ReportedOct 2026

Global PIQAglobal-piqa · reasoning

89.8%

ReportedOct 2026

GPQAgpqa · reasoning

88.4%

ReportedOct 2026

Humanity's Last Examhumanity's-last-exam · reasoning

28.7%

ReportedOct 2026

Learn what each benchmark measures ↗

Pricing detail

Input price per 1M tokens: $0.75
Output price per 1M tokens: $4.50
Blended price per 1M tokens (50/50): $2.63

Prices are published by Venice AI. Venice model page ↗

Capabilities

ReasoningFunction callingVisionVideo inputWeb search

Endpoint metadata

Privacy: anonymized
Quantization: not-available
Maximum completion tokens: 33K
Uncensored: No
Beta: No
Mapping confidence: 100.0%

View model source ↗

All capabilities

Max Images: 5
Quantization: not-available
Supports E2 E E: No
Supports Vision: Yes
Supports X Search: No
Optimized For Code: Yes
Supports Log Probs: Yes
Supports Reasoning: Yes
Supports Web Search: Yes
Supports Audio Input: No
Supports Video Input: Yes
Default Reasoning Effort: low
Reasoning Effort Options: none, low, medium, high
Supports Multiple Images: Yes
Supports Response Schema: Yes
Supports Tee Attestation: No
Supports Function Calling: Yes
Supports Reasoning Effort: Yes

Related models

Qwen 3.5 397B

Benchmark results

More from qwen