Qwen
·Released on Jul 21, 2025Qwen3 235B A22B Instruct 2507 API Benchmarks, Pricing & Provider Data
Compare Qwen3 235B A22B Instruct 2507 with another model
Choose a model to open its comparison page.
LLM
Qwen3-235B-A22B-Instruct-2507 is a multilingual, instruction-tuned mixture-of-experts language model based on the Qwen3-235B architecture, with 22B active parameters per forward pass. It is optimized ...
Category Performance
Observed-capability estimates with uncertainty reported separately from independent benchmark families.
- Coverage
- 3 / 8
- Methodology
- V3.0
#1Math62.9Estimated2/4 Measured dimensions
80% interval: 51.1–74.7
foundational math62.1
math_500 · math_500
competition math63.7
aime · aime
proof frontierPrior only
applied tool mathPrior only
#2Coding55.5Provisional1/4 Measured dimensions
80% interval: 41.5–69.4
code generation55.5
livecodebench · livecodebench / scicode · scicode
repository engineeringPrior only
debugging testingPrior only
tooling qualityPrior only
#3Reasoning53.6Estimated2/4 Measured dimensions
80% interval: 42.5–64.7
abstract logicPrior only
scientific causal51.8
gpqa · gpqa / hle · hle
multistep constraints55.5
mmlu_pro · mmlu_pro
evidence verificationPrior only
No data:AgentsKnowledgeMultilingualMultimodalInstruction following
Specifications
Input and output token limits for this model, plus how it ranks on long-context understanding.
INPUT
262.1Ktokens
≈ 314.6 pages of text
OUTPUT
16.4Ktokens
8K128K1M4M
262.1K
Features
Technical Details
- Input
- Output
- Total parameters
- 235B
- Active parameters
- 22B
- Released
- Jul 2025
- Knowledge cutoff
- 2025-06-30
- Tokenizer
- Qwen3
- Architecture
- text->text
- Moderated
- No
- Supported parameters
- frequency_penaltylogit_biaslogprobsmax_tokensmin_ppresence_penaltyrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p
Rankings
Excels at
Detailed scores
Updated: Aug 23, 2026Pricing
undefined metric} other undefined metrics}}
Input price$0.230/M#42 / 180Output price$2.30/M#82 / 180
Pricing
undefined metric} other undefined metrics}}
Benchmark
LMSpeed rank
Score
Source
Updated
Confidence
Agents
V3.0undefined metric} other undefined metrics}} · No data
No data80% interval30.8–69.20/4 Measured dimensions
Agents
V3.0undefined metric} other undefined metrics}} · No data
Dimensions and evidence
Planning & decompositionPrior only
Tool usePrior only
Environment & long-horizon executionPrior only
Recovery & completion reliabilityPrior only
Coding
V3.0undefined metric} other undefined metrics}} · Provisional
Score55.580% interval41.5–69.41/4 Measured dimensionsLiveCodeBench78.8%#21 / 115SciCode42.4%#62 / 204
Coding
V3.0undefined metric} other undefined metrics}} · Provisional
Dimensions and evidence
Code generation55.5
undefined family} other undefined families}} · undefined metric} other undefined metrics}}
livecodebench · livecodebench · z 0.90 · q 1.00
scicode · scicode · z 0.39 · q 1.00
Repository engineeringPrior only
Debugging & testingPrior only
Tool-assisted development & qualityPrior only
Benchmark
LMSpeed rank
Score
Source
Updated
Confidence
LiveCodeBenchV3 evidence
LMSpeed rank#21
Score78.8%
Sourceartificialanalysis.ai
UpdatedAug 23, 2026
Confidence4
SciCodeV3 evidence
LMSpeed rank#62
Score42.4%
Sourceartificialanalysis.ai
UpdatedAug 23, 2026
Confidence4
Reasoning
V3.0undefined metric} other undefined metrics}} · Estimated
Score53.680% interval42.5–64.72/4 Measured dimensionsMMLU-Pro84.3%#33 / 129GPQA79.0%#94 / 211
Reasoning
V3.0undefined metric} other undefined metrics}} · Estimated
Dimensions and evidence
Abstract logicPrior only
Scientific & causal reasoning51.8
undefined family} other undefined families}} · undefined metric} other undefined metrics}}
gpqa · gpqa · z 0.11 · q 1.00
hle · hle · z 0.11 · q 1.00
Multi-step constraints55.5
undefined family} other undefined families}} · undefined metric} other undefined metrics}}
mmlu_pro · mmlu_pro · z 0.53 · q 1.00
Evidence integration & verificationPrior only
Benchmark
LMSpeed rank
Score
Source
Updated
Confidence
MMLU-ProV3 evidence
LMSpeed rank#33
Score84.3%
Sourceartificialanalysis.ai
UpdatedAug 23, 2026
Confidence4
Knowledge
V3.0undefined metric} other undefined metrics}} · No data
No data80% interval30.8–69.20/4 Measured dimensions
Knowledge
V3.0undefined metric} other undefined metrics}} · No data
Dimensions and evidence
Broad knowledgePrior only
Professional knowledgePrior only
FactualityPrior only
Retrieval & open-book usePrior only
Math
V3.0undefined metric} other undefined metrics}} · Estimated
Score62.980% interval51.1–74.72/4 Measured dimensionsAIME94.0%#1 / 68MATH-50098.4%#5 / 73
Math
V3.0undefined metric} other undefined metrics}} · Estimated
Dimensions and evidence
Foundational math62.1
undefined family} other undefined families}} · undefined metric} other undefined metrics}}
math_500 · math_500 · z 1.65 · q 1.00
Competition math63.7
undefined family} other undefined families}} · undefined metric} other undefined metrics}}
aime · aime · z 1.80 · q 1.00
Advanced proofsPrior only
Applied & tool-assisted mathPrior only
Benchmark
LMSpeed rank
Score
Source
Updated
Confidence
MATH-500V3 evidence
LMSpeed rank#5
Score98.4%
Sourceartificialanalysis.ai
UpdatedAug 23, 2026
Confidence4
Multilingual
V3.0undefined metric} other undefined metrics}} · No data
No data80% interval30.8–69.20/4 Measured dimensions
Multilingual
V3.0undefined metric} other undefined metrics}} · No data
Dimensions and evidence
Cross-language understandingPrior only
Multilingual generationPrior only
Reasoning transferPrior only
Low-resource robustnessPrior only
Multimodal
V3.0undefined metric} other undefined metrics}} · No data
No data80% interval30.8–69.20/4 Measured dimensions
Multimodal
V3.0undefined metric} other undefined metrics}} · No data
Dimensions and evidence
Perception & OCRPrior only
Document & spatial understandingPrior only
Visual reasoningPrior only
Video & grounded actionPrior only
Instruction following
V3.0undefined metric} other undefined metrics}} · No data
No data80% interval30.8–69.20/4 Measured dimensions
Instruction following
V3.0undefined metric} other undefined metrics}} · No data
Dimensions and evidence
Constraint followingPrior only
Structured outputPrior only
Novel-instruction generalizationPrior only
Multi-turn & long instructionsPrior only
OpenRouter endpoints
10 endpointsThird-party OpenRouter endpoint data, shown separately from LMSpeed measurements. Some 30-minute live performance fields only appear after syncing with an OpenRouter API key.
| Provider endpoint | Input | Output | 1d uptime | 30m latency | 30m throughput | Context / output |
|---|---|---|---|---|---|---|
Google google-vertex/us-south1 | $0.220/M | $0.880/M | 100.0% | — | — | undefined tokens / undefined tokens |
Alibaba alibaba | $0.150/M | $0.598/M | 99.9% | — | — | undefined tokens / undefined tokens |
Parasail parasail/fp8 | $0.140/M | $0.800/M | 99.9% | — | — | undefined tokens / undefined tokens |
StreamLake streamlake | $0.210/M | $0.840/M | 99.7% | — | — | undefined tokens / undefined tokens |
Novita novita/fp8 | $0.090/M | $0.580/M | 99.6% | — | — | undefined tokens / undefined tokens |
GMICloud gmicloud/fp8 | $0.087/M | $0.350/M | 99.5% | — | — | undefined tokens / — |
AtlasCloud atlas-cloud/fp8 | $0.200/M | $0.880/M | 99.5% | — | — | undefined tokens / undefined tokens |
DeepInfra deepinfra/fp8 | $0.090/M | $0.550/M | 99.1% | — | — | undefined tokens / undefined tokens |
Venice venice/fp8 | $0.150/M | $0.750/M | 98.2% | — | — | undefined tokens / undefined tokens |
Nebius nebius/fp8 | $0.200/M | $0.600/M | 98.1% | — | — | undefined tokens / — |