Back to Developer ZoneOfficial Pricing Official Pricing Official Pricing Official Pricing Official Pricing Official Pricing Official Pricing Official Pricing Official Pricing Official Pricing
Alibaba Cloud Models
Explore all 11 models from Alibaba Cloud with detailed pricing, pros & cons, and developer recommendations.
11
Models
$0.0000
Lowest Input
1M
Max Context
3
Quality Tiers
Quick Recommendations
Best Value: Qwen-RobotWorld ($0.0000/1M)
Best Quality: Qwen3.7-Max
Qwen3.7-Max
FlagshipLong-horizon agent workflows, coding agents, complex reasoning
When to use: Frontier agent workloads requiring long autonomous runs, complex multi-step coding tasks, and deep research analysis.
Upgrade Highlights
- ◆1M token context — removes limits on document-heavy agent work
- ◆65K max output — massive single-turn generation
- ◆Sustained 35-hour autonomous kernel optimization (1,158 tool calls)
- ◆SWE-Verified 80.4, LiveCodeBench 91.6 — rivals Claude Opus 4.6
- ◆OpenAI + Anthropic API compatible — drop-in replacement
Input Price
$2.50
per 1M tokens
Output Price
$7.50
per 1M tokens
Cached Input
$0.250
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 1M
Max Output: 65,536 tokens
Knowledge Cutoff: 2026-05
VisionFunction CallingFine-tuningJSON Mode
Pros
- 1M context window for document-heavy agent work
- 65K max output — longest in Qwen family
- Cross-harness compatibility (Claude Code, OpenClaw, Qwen Code)
- 35-hour sustained autonomous execution
- Competitive with Claude Opus 4.6 on coding benchmarks
Cons
- Proprietary — no open weights or self-hosting
- Higher cost than Qwen 3.6 line
- No vision support
- API-only access
Performance
Output Speed~55 tok/s
Rate Limit2,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU-Pro
89.6%
LiveCodeBench
91.6%
SWE-Verified
80.4%
GPQA Diamond
92.4%
Agents Using This Model
3Qwen3.7-Plus
Mid-tierMultimodal tasks, cost-effective agent deployment
When to use: Cost-effective multimodal deployments needing video and image understanding alongside text, with long context requirements.
Upgrade Highlights
- ◆Multimodal input: text + video + image in one model
- ◆1M context at $0.40/1M — 6x cheaper than Qwen3.7-Max
- ◆Strong agent capability at mid-tier cost
- ◆OpenAI-compatible API
Input Price
$0.400
per 1M tokens
Output Price
$1.60
per 1M tokens
Cached Input
$0.100
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 1M
Max Output: 16,384 tokens
Knowledge Cutoff: 2026-05
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- 1M context at mid-tier pricing
- Multimodal: text, video, and image input
- Strong speed-capability balance
- Proprietary but very affordable
Cons
- Proprietary — no self-hosting
- Less capable than Qwen3.7-Max on complex reasoning
- 16K max output
Performance
Output Speed~80 tok/s
Rate Limit5,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU-Pro
84.2%
LiveCodeBench
78.5%
MMMU
72.1%
Qwen3-235B-A22B
FlagshipComplex reasoning, multilingual tasks
When to use: Best value flagship for multilingual workloads, complex reasoning, and cost-sensitive production deployments.
Upgrade Highlights
- ◆MoE architecture: 235B params, only 22B active — GPT-4 class at 1/10 the price
- ◆131K context — handles long documents and codebases
- ◆100+ language support — best-in-class for non-English tasks
- ◆Open-source: full weights on HuggingFace for self-hosting
- ◆$0.40/$1.20 per 1M tokens — undercuts GPT-4o by 90%
Input Price
$0.400
per 1M tokens
Output Price
$1.20
per 1M tokens
Cached Input
$0.100
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 131K
Max Output: 8,192 tokens
Knowledge Cutoff: 2025-04
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- MoE 235B total / 22B active — flagship performance at low cost
- 131K context window
- Strong multilingual (100+ languages)
- Open-source weights available
Cons
- No vision support
- Max output 8K tokens
- Less ecosystem integration than GPT-4
Performance
Output Speed~70 tok/s
Rate Limit5,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU
86.8%
LiveCodeBench
63.7%
IFEval
86.2%
Agents Using This Model
2Qwen3-30B-A3B
Mid-tierEfficient multilingual inference
When to use: High-throughput multilingual tasks where cost efficiency matters most.
Upgrade Highlights
- ◆Only 3B active params — runs on consumer GPUs
- ◆131K context at $0.15/1M input — cheapest long-context option
- ◆Open-source for full customization
- ◆Strong function calling for agent workflows
Input Price
$0.150
per 1M tokens
Output Price
$0.600
per 1M tokens
Cached Input
$0.040
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 131K
Max Output: 8,192 tokens
Knowledge Cutoff: 2025-04
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- MoE 30B total / 3B active — ultra-efficient
- 131K context
- Excellent cost-performance ratio
- Open-source
Cons
- Smaller active params limit complex reasoning
- No vision
- 8K max output
Performance
Output Speed~120 tok/s
Rate Limit10,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU
78.5%
LiveCodeBench
48.2%
Qwen3-32B
Mid-tierBalanced performance and cost
When to use: When you need reliable dense model performance for coding and general tasks.
Upgrade Highlights
- ◆Dense 32B architecture — no MoE routing overhead
- ◆131K context for long-form content
- ◆Strong coding: LiveCodeBench 55.3%
- ◆Open-source with full HuggingFace support
Input Price
$0.200
per 1M tokens
Output Price
$0.600
per 1M tokens
Cached Input
$0.050
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 131K
Max Output: 8,192 tokens
Knowledge Cutoff: 2025-04
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- Dense 32B — consistent performance
- 131K context
- Strong coding ability
- Open-source
Cons
- No vision
- 8K max output
- Higher latency than MoE variants
Performance
Output Speed~65 tok/s
Rate Limit5,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU
83.2%
LiveCodeBench
55.3%
Qwen3-14B
LiteLightweight general tasks
When to use: Budget-friendly option for summarization, translation, and simple Q&A.
Upgrade Highlights
- ◆14B dense — fits on single GPU
- ◆131K context at just $0.10/1M input
- ◆Good enough for most everyday tasks
- ◆Open-source for fine-tuning
Input Price
$0.100
per 1M tokens
Output Price
$0.300
per 1M tokens
Cached Input
$0.030
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 131K
Max Output: 8,192 tokens
Knowledge Cutoff: 2025-04
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- Compact 14B dense model
- 131K context
- Very low cost
- Open-source
Cons
- Limited complex reasoning
- No vision
- 8K max output
Performance
Output Speed~90 tok/s
Rate Limit10,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU
77.1%
LiveCodeBench
42.8%
Qwen3-8B
LiteEdge deployment, simple tasks
When to use: Edge devices, local deployment, or ultra-low-cost batch processing.
Upgrade Highlights
- ◆8B params — runs on RTX 3060 or equivalent
- ◆$0.05/1M input — among the cheapest available
- ◆131K context despite small size
- ◆Ideal for local/offline deployment
Input Price
$0.050
per 1M tokens
Output Price
$0.150
per 1M tokens
Cached Input
$0.010
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 131K
Max Output: 8,192 tokens
Knowledge Cutoff: 2025-04
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- Tiny 8B — runs on laptop GPUs
- 131K context
- Extremely cheap
- Open-source
Cons
- Basic reasoning only
- No vision
- 8K max output
Performance
Output Speed~150 tok/s
Rate Limit20,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMLU
71.5%
LiveCodeBench
33.1%
Qwen-VL-Plus
Mid-tierMultimodal understanding, document analysis
When to use: Document analysis, image captioning, visual Q&A, and multimodal RAG pipelines.
Upgrade Highlights
- ◆Native multimodal — processes images and text together
- ◆131K context handles multi-page documents
- ◆Strong OCR: chart, table, and diagram understanding
- ◆Multilingual VQA across 100+ languages
Input Price
$0.200
per 1M tokens
Output Price
$0.800
per 1M tokens
Cached Input
$0.050
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 131K
Max Output: 8,192 tokens
Knowledge Cutoff: 2025-04
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- Native vision-language model
- 131K context with images
- Strong document OCR and chart understanding
- Multilingual VQA
Cons
- No fine-tuning
- 8K max output
- Higher cost than text-only Qwen3
Performance
Output Speed~55 tok/s
Rate Limit3,000 RPM
Multimodal
Image InputImage OutputAudio InputAudio Output
Benchmarks
MMMU
68.2%
MathVista
62.5%
Qwen-RobotManip
FlagshipRobotic manipulation, dexterous hand control
When to use: For robotic manipulation tasks: grasping, assembly, and dexterous hand control in research and industrial settings.
Upgrade Highlights
- ◆First Qwen-Robot VLA manipulation model
- ◆38,100+ hours of open-source training data
- ◆Unified state-action space across robot types
- ◆Camera-frame end-effector incremental pose control
- ◆Part of complete Qwen-Robot Suite (Manip + Nav + World)
Input Price
$0.0000
per 1M tokens
Output Price
$0.0000
per 1M tokens
Cached Input
—
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 0
Max Output: 0 tokens
Knowledge Cutoff: 2026-06
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- VLA model for precise robotic manipulation
- 38,100+ hours of training from open-source data
- Multi-robot-type support via unified action space
- Open-source under Apache 2.0
Cons
- Specialized for robotics — not a general LLM
- Requires robot hardware or simulator for deployment
- No text generation capabilities
- Very new — limited community adoption
Performance
Output Speed—
Rate Limit—
Multimodal
Image InputImage OutputAudio InputAudio Output
Qwen-RobotWorld
FlagshipPhysical world prediction, robot planning
When to use: For robot planning and world simulation: predicting outcomes of actions across manipulation, driving, and navigation scenarios.
Upgrade Highlights
- ◆World model: predicts physically plausible futures
- ◆Cross-scene: works across manipulation, driving, navigation
- ◆Natural language action interface
- ◆Open-source: full weights for research and deployment
- ◆Part of complete Qwen-Robot Suite (Manip + Nav + World)
Input Price
$0.0000
per 1M tokens
Output Price
$0.0000
per 1M tokens
Cached Input
—
per 1M tokens
Batch Input
—
per 1M tokens
Context Window: 0
Max Output: 0 tokens
Knowledge Cutoff: 2026-06
VisionFunction CallingFine-tuningJSON ModeFree Tier
Pros
- World model for predicting physically plausible futures
- Cross-scene: manipulation, driving, and navigation
- Natural language action interface for intuitive control
- Open-source under Apache 2.0
Cons
- Specialized for world simulation only
- No text generation or robot control
- Requires integration with Manip/Nav for full stack
- Very new — limited benchmarks available
Performance
Output Speed—
Rate Limit—
Multimodal
Image InputImage OutputAudio InputAudio Output
Side-by-Side Comparison
| Model | Tier | Input | Output | Context |
|---|---|---|---|---|
| Qwen3.7-Max | Flagship | $2.50 | $7.50 | 1M |
| Qwen3.7-Plus | Mid-tier | $0.400 | $1.60 | 1M |
| Qwen3-235B-A22B | Flagship | $0.400 | $1.20 | 131K |
| Qwen3-30B-A3B | Mid-tier | $0.150 | $0.600 | 131K |
| Qwen3-32B | Mid-tier | $0.200 | $0.600 | 131K |
| Qwen3-14B | Lite | $0.100 | $0.300 | 131K |
| Qwen3-8B | Lite | $0.050 | $0.150 | 131K |
| Qwen-VL-Plus | Mid-tier | $0.200 | $0.800 | 131K |
| Qwen-RobotManip | Flagship | $0.0000 | $0.0000 | 0 |
| Qwen-RobotNav | Flagship | $0.0000 | $0.0000 | 0 |
| Qwen-RobotWorld | Flagship | $0.0000 | $0.0000 | 0 |