moonshotai/kimi-k2-thinking-20251106
Kimi K2 Thinking is Moonshot AI’s most advanced open reasoning model to date, extending the K2 series into agentic, long-horizon reasoning. Built on the trillion-parameter Mixture-of-Experts (MoE) architecture introduced in Kimi K2, it activates 32 billion parameters per forward
- Context window
- 262,144
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 22
- OpenCompass LLM: 94.1%
qwen/qwen3.5-9b-20260310
Qwen3.5-9B is a multimodal foundation model from the Qwen3.5 family, designed to deliver strong reasoning, coding, and visual understanding in an efficient 9B-parameter architecture. It uses a unified vision-language design with early fusion of multimodal tokens, allowing the mod
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 13.3
- OpenRouter Benchmarks: 74.405%
Inference prices
- Input tokens (per 1M) : $0.10 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.15 / output_tokens_per_million · siliconFlow ↗
meta-llama/llama-3.3-70b-instruct
The Meta Llama 3.3 multilingual large language model (LLM) is a pretrained and instruction tuned generative model in 70B (text in/text out). The Llama 3.3 instruction tuned text only model is optimized for multilingual dialogue use cases and outperforms many of the available open
- Context window
- 131,072
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 7.7
- Open LLM Leaderboard v2: 44.847%
- Open Persian LLM Leaderboard: 71.4%
- OpenRouter Benchmarks: 49.952%
qwen/qwen3-235b-a22b-thinking-2507
Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) language model optimized for complex reasoning tasks. It activates 22B of its 235B parameters per forward pass and natively supports up to 262,144 tokens of context. This "thinking-only" var
- Context window
- 128,000
- Input modalities
- text
- Output modalities
- text
- Capabilities
- reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 12.7
- LMArena: 1,415.543
- OpenCompass LLM: 83.5%
- OpenRouter Benchmarks: 80.303%
openai/gpt-oss-20b
gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference and deployability on consumer or single-GPU hardw
- Context window
- 131,072
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 10
- HELM Capabilities: 56.467%
- Kaggle Benchmarks: 100%
- LLM Stats · GitHub archive: 0.173
- OpenCompass LLM: 87.9%
- OpenRouter Benchmarks: 44.877%
Inference prices
- Input tokens (per 1M) : $0.04 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.18 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3-32b-04-28
Qwen3-32B is a dense 32.8B parameter causal language model from the Qwen3 series, optimized for both complex reasoning and efficient dialogue. It supports seamless switching between a "thinking" mode for tasks like math, coding, and logical inference, and a "non-thinking" mode fo
- Context window
- 131,072
- Input modalities
- text
- Output modalities
- text
- Capabilities
- reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 7.3
- BFCL: 48.71%
- LLM Stats · GitHub archive: 0.73
- LMArena: 1,340.076
- OpenCompass LLM: 57.6%
- OpenRouter Benchmarks: 63.343%
Inference prices
- Input tokens (per 1M) : $0.14 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.57 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3-14b-04-28
Qwen3-14B is a dense 14.8B parameter causal language model from the Qwen3 series, designed for both complex reasoning and efficient dialogue. It supports seamless switching between a "thinking" mode for tasks like math, programming, and logical inference, and a "non-thinking" mod
- Context window
- 40,960
- Input modalities
- text
- Output modalities
- text
- Capabilities
- reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 6.7
- BFCL: 41.03%
- OpenRouter Benchmarks: 59.788%
Inference prices
- Input tokens (per 1M) : $0.07 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.28 / output_tokens_per_million · siliconFlow ↗
meta-llama/llama-3.1-8b-instruct
Meta's latest class of model (Llama 3.1) launched with a variety of sizes & flavors. This 8B instruct-tuned version is fast and efficient. It has demonstrated strong performance compared to...
- Context window
- 131,072
- Input modalities
- text
- Output modalities
- text
- Capabilities
- structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 6.9
- Open LLM Leaderboard v2: 23.764%
- OpenRouter Benchmarks: 29.121%
deepseek/deepseek-v3.2-exp
DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to improve training and inference effici
- Context window
- 163,840
- Input modalities
- text
- Output modalities
- text
- Capabilities
- reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 16.6
- BFCL: 56.73%
- LMArena: 1,422.402
Inference prices
- Input tokens (per 1M) : $0.27 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.41 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3-235b-a22b-07-25
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 for general-purpose text generation, including instruction following, logical re
- Context window
- 262,144
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 12
- LLM Stats · GitHub archive: 0.95
- LMArena: 1,419.17
- OpenCompass LLM: 12.3%
- OpenRouter Benchmarks: 40.667%
stepfun/step-3.5-flash
Step 3.5 Flash is StepFun's most capable open-source foundation model. Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token....
- Context window
- 262,144
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, streaming, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 17
- LMArena: 1,402.661
- OpenCompass LLM: 83.9%
Inference prices
- Input tokens (per 1M) : $0.10 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.30 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3-30b-a3b-instruct-2507
Qwen3-30B-A3B-Instruct-2507 is a 30.5B-parameter mixture-of-experts language model from Qwen, with 3.3B active parameters per inference. It operates in non-thinking mode and is designed for high-quality instruction following, multilingual understanding, and agentic tool use. Post
- Context window
- 262,144
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 7.5
- BFCL: 41.39%
- LMArena: 1,383.838
- OpenCompass LLM: 7.7%
- OpenRouter Benchmarks: 64.141%
Inference prices
- Input tokens (per 1M) : $0.09 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.30 / output_tokens_per_million · siliconFlow ↗
vertexAI/publishers/qwen/models/qwq@qwq-32b
- Input modalities
- text
- Output modalities
- text
- Capabilities
- toolCalling
Benchmarks
- LiveBench: 60.788%
- LLM Stats · GitHub archive: 0.906
- LMArena: 1,329.075
- Open LLM Leaderboard v2: 12.215%
qwen/qwen-2.5-72b-instruct
Qwen2.5 72B is the latest series of Qwen large language models. Qwen2.5 brings the following improvements upon Qwen2: - Significantly more knowledge and has greatly improved capabilities in coding and...
- Context window
- 32,768
- Input modalities
- text
- Output modalities
- text
- Capabilities
- structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 7.7
- LMArena: 1,269.026
- Open LLM Leaderboard v2: 47.98%
- OpenRouter Benchmarks: 27.333%
Inference prices
- Input tokens (per 1M) : $0.59 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.59 / output_tokens_per_million · siliconFlow ↗
qwen/qwen-2.5-coder-32b-instruct
Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (formerly known as CodeQwen). Qwen2.5-Coder brings the following improvements upon CodeQwen1.5: - Significantly improvements in **code generation**, **code reasoning**...
- Context window
- 32,768
- Input modalities
- text
- Output modalities
- text
- Capabilities
- structuredOutputs
Benchmarks
- Artificial Analysis: 6.7
- BigCode Models Leaderboard: 83.2%
- LMArena: 1,230.051
- Open LLM Leaderboard v2: 39.885%
- SWE-bench: 38%
siliconFlow/baidu/ERNIE-4.5-300B-A47B
ERNIE-4.5-300B-A47B is a large language model developed by Baidu based on a Mixture-of-Experts (MoE) architecture. The model has a total of 300 billion parameters, but only activates 47 billion parameters per token during inference, thus balancing powerful performance with comput
- Maximum output
- 131,072
- Input modalities
- text
- Output modalities
- text
- Capabilities
- structuredOutputs
siliconFlow/black-forest-labs/FLUX-1.1-pro
FLUX1.1 Pro is an enhanced text-to-image model built on the FLUX.1 architecture, offering improved composition, detail, and rendering speed. With better visual consistency and artistic fidelity, it's suitable for illustration, creative content generation, and e-commerce visual as
- Input modalities
- text
- Output modalities
- image
- Capabilities
- imageOutput
Inference prices
siliconFlow/black-forest-labs/FLUX-1.1-pro-Ultra
FLUX1.1 Pro Ultra is the high-resolution version of FLUX1.1 Pro, capable of generating images up to 4 megapixels (2K resolution). It improves photo realism and prompt controllability for advanced use cases. The Ultra mode is optimized for composition and precision, while Raw mode
- Input modalities
- text
- Output modalities
- image
- Capabilities
- imageOutput
Inference prices