Model comparison
Kimi-VL-A3B-Thinking-2506 vs Qwen2.5-VL
- Input
- $0.025 / 1M tokens
- Output
- $0.10 / 1M tokens
- Context
- 128K
- Released
- 2025-06-21
- Input
- $0.80 per 1M input tokens
- Output
- $1.00 per 1M output tokens
- Context
- 32,768 tokens (extendable to ~64K for long video via YaRN; some API providers expose up to 131K)
- Released
- 2025-01-26
Qwen2.5-VL scores higher on 4 of the 7 benchmarks both models report (the other leads on 3); Kimi-VL-A3B-Thinking-2506 is cheaper on input tokens ($0 vs $0.8 per 1M, 32×); Kimi-VL-A3B-Thinking-2506 has the larger context window (128K vs 33K).
Specs and pricing side by side
| Spec | Kimi-VL-A3B-Thinking-2506 | Qwen2.5-VL |
|---|---|---|
| Maker | Moonshot AI (Kimi) | Alibaba (Qwen) |
| Released | 2025-06-21 | 2025-01-26 |
| Context window | 128K | 32,768 tokens (extendable to ~64K for long video via YaRN; some API providers expose up to 131K) |
| Max output | 32K | Not separately documented; bounded by the… |
| Input price | $0.025 / 1M tokens | $0.80 per 1M input tokens |
| Output price | $0.10 / 1M tokens | $1.00 per 1M output tokens |
| Parameters | 16B total / 2.8B active (MoE) | Four open sizes: 3B, 7B, 32B and 72B (each in base and instruction-tuned variants) |
| License | MIT | Mixed by size: Qwen2.5-VL-72B-Instruct under the Qwen License; 7B and 32B under Apache 2.0; 3B under the Qwen Research License (non-commercial only). |
| Weights | Open weights | Open weightssame |
| Modalities | Text, Vision, Video, PDF | text, image, video |
| Knowledge cutoff | December 2024 | Not officially published by Alibaba |
Benchmarks both models report
| Benchmark | Kimi-VL-A3B-Thinking-2506 | Qwen2.5-VL |
|---|---|---|
| MathVision | 56.9% | 38.1% |
| MathVista | 80.1% | 74.8% |
| MMMU | 64% | 70.2% |
| MMMU-Pro | 46.3% | 51.1% |
| MMStar | 70.4% | 70.8% |
| Video-MME | 71.9% | 73.3% |
| ScreenSpot-Pro | 52.8% | 43.6% |
Scores are as published by each maker or its cited source; harness, reasoning effort and date can differ between rows, so treat small gaps as ties. Sources are listed on each model page.
Which one should you pick?
Pick Kimi-VL-A3B-Thinking-2506 if…
- You care about MathVision, MathVista, ScreenSpot-Pro (it scores higher on 3 shared benchmarks).
- Cost matters: input tokens cost 32× less ($0 vs $0.8 per 1M).
- You need the longer context (128K vs 33K).
- You need Text / Vision / Video / PDF input, which Qwen2.5-VL lacks.
- Your use case is: visual and multimodal math/reasoning over charts, diagrams and screenshots.
Pick Qwen2.5-VL if…
- You care about MMMU, MMMU-Pro, MMStar (it scores higher on 4 shared benchmarks).
- You need text / image / video input, which Kimi-VL-A3B-Thinking-2506 lacks.
- Your use case is: document AI: extracting text, tables and layout from invoices, contracts, research papers and scanned forms.
FAQ
Which is better, Kimi-VL-A3B-Thinking-2506 or Qwen2.5-VL?
Qwen2.5-VL scores higher on 4 of the 7 benchmarks both models report (the other leads on 3); Kimi-VL-A3B-Thinking-2506 is cheaper on input tokens ($0 vs $0.8 per 1M, 32×); Kimi-VL-A3B-Thinking-2506 has the larger context window (128K vs 33K). The better pick depends on whether you value benchmark scores, price per token, context length or openness for your workload.
How much cheaper is Kimi-VL-A3B-Thinking-2506?
Kimi-VL-A3B-Thinking-2506: $0.025 / 1M tokens input, $0.10 / 1M tokens output. Qwen2.5-VL: $0.80 per 1M input tokens input, $1.00 per 1M output tokens output (list API prices as published by each maker).
What are the context windows of Kimi-VL-A3B-Thinking-2506 and Qwen2.5-VL?
Kimi-VL-A3B-Thinking-2506 offers 128K with up to 32K output; Qwen2.5-VL offers 32,768 tokens (extendable to ~64K for long video via YaRN; some API providers expose up to 131K) with up to Not separately documented; bounded by the 32,768-token context window output.
Is Kimi-VL-A3B-Thinking-2506 newer than Qwen2.5-VL?
Kimi-VL-A3B-Thinking-2506 was released 2025-06-21; Qwen2.5-VL was released 2025-01-26.
Go deeper
- Kimi-VL-A3B-Thinking-2506: full specs, pricing and benchmarks
- Qwen2.5-VL: full specs, pricing and benchmarks
- qwen3 vl vs qwen2 5 vl
- qwen3 vl vs kimi vl a3b thinking 2506
- glm 4 1v thinking vs kimi vl a3b thinking 2506
- qwen2 5 vl vs pixtral 12b
- mistral small 3 1 vs qwen2 5 vl
- glm 4 5v vs kimi vl a3b thinking 2506
- All model comparisons