deepseek/deepseek-v4-flash-20260731
DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows. This is the GA release of DeepSeek V4 Flash.
- Context window
- 1,048,576
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 34.3
- Kaggle Benchmarks: 1,472.3
- OpenRouter Benchmarks: 88.72%
- Vercel AI Gateway: 10.855%
Inference prices
- Cached input tokens (per 1M) : $0.01 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.22 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.66 / output_tokens_per_million · siliconFlow ↗
z-ai/glm-5.2-20260616
GLM 5.2 is a large-scale reasoning model from Z.ai. It supports text input and output with a 1M-token context window, and is suited for long-horizon agent workflows, project-level software engineering, and complex multi-step automation. Reasoning efforts `high` and `xhigh` are s
- Context window
- 1,048,576
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 33.7
- OpenRouter Benchmarks: 76.933%
Inference prices
- Cached input tokens (per 1M) : $0.26 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $1.30 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $4.09 / output_tokens_per_million · siliconFlow ↗
moonshotai/kimi-k2.5-0127
Kimi K2.5 is Moonshot AI's native multimodal model, delivering state-of-the-art visual coding capability and a self-directed agent swarm paradigm. Built on Kimi K2 with continued pretraining over approximately 15T mixed visual and text tokens, it delivers strong performance in ge
- Context window
- 262,144
- Input modalities
- text, image
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 23.5
- OpenRouter Benchmarks: 90.284%
Inference prices
- Cached input tokens (per 1M) : $0.07 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.45 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $2.25 / output_tokens_per_million · siliconFlow ↗
z-ai/glm-4.6
Compared with GLM-4.5, this generation brings several key improvements: Longer context window: The context window has been expanded from 128K to 200K tokens, enabling the model to handle more complex...
- Context window
- 204,800
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 18.5
- BFCL: 72.38%
- OpenCompass LLM: 78.2%
- OpenRouter Benchmarks: 85.607%
- SWE-bench: 55.4%
z-ai/glm-4.7-20251222
GLM-4.7 is Z.ai’s latest flagship model, featuring upgrades in two key areas: enhanced programming capabilities and more stable multi-step reasoning/execution. It demonstrates significant improvements in executing complex agent tasks while delivering more natural conversational e
- Context window
- 204,800
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 22.2
- OpenCompass LLM: 83.8%
- OpenRouter Benchmarks: 83.694%
Inference prices
- Input tokens (per 1M) : $0.42 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $2.20 / output_tokens_per_million · siliconFlow ↗
deepseek/deepseek-v3.2-20251201
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism that reduces training and inference cost wh
- Context window
- 163,840
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 21.5
- OpenCompass LLM: 85.8%
- OpenRouter Benchmarks: 72.8%
Inference prices
- Cached input tokens (per 1M) : $0.14 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.27 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.42 / output_tokens_per_million · siliconFlow ↗
tencent/hy3-preview-20260421
Hy3 preview is a high-efficiency Mixture-of-Experts model from Tencent designed for agentic workflows and production use. It supports configurable reasoning levels across disabled, low, and high modes, allowing it to balance speed and depth depending on the task, while delivering
- Context window
- 262,144
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, toolCalling
Benchmarks
- Artificial Analysis: 25.3
- LMArena: 1,401.297
- OpenCompass LLM: 81%
moonshotai/kimi-k2.6-20260420
Kimi K2.6 is Moonshot AI's next-generation multimodal model, designed for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration. It handles complex end-to-end coding tasks across Python, Rust, and Go, and can convert prompts and visual inputs into pro
- Context window
- 262,144
- Input modalities
- text, image
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, videoInput, visionInput
Benchmarks
- Artificial Analysis: 27
- OpenCompass LLM: 82.7%
- OpenRouter Benchmarks: 89.625%
Inference prices
- Cached input tokens (per 1M) : $0.14 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.77 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $3.40 / output_tokens_per_million · siliconFlow ↗
moonshotai/kimi-k2.7-code-20260612
MoonshotAI: Kimi K2.7 Code is a coding-focused model in Moonshot AI's Kimi K2 family, built to complete end-to-end programming tasks reliably over long contexts. It uses a native multimodal mixture-of-experts architecture that accepts text and image input, and it always operates
- Context window
- 262,144
- Input modalities
- text, image
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, videoInput, visionInput
Benchmarks
- Artificial Analysis: 25.8
- OpenRouter Benchmarks: 74.333%
Inference prices
- Cached input tokens (per 1M) : $0.18 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.86 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $3.80 / output_tokens_per_million · siliconFlow ↗
z-ai/glm-5.1-20260406
GLM-5.1 delivers a major leap in coding capability, with particularly significant gains in handling long-horizon tasks. Unlike previous models built around minute-level interactions, GLM-5.1 can work independently and continuously on a single task for more than 8 hours, autonomou
- Context window
- 204,800
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 26.1
- OpenRouter Benchmarks: 83.237%
Inference prices
- Cached input tokens (per 1M) : $0.60 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $1.19 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $3.74 / output_tokens_per_million · siliconFlow ↗
deepseek/deepseek-v4-flash-20260423
DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window. It is designed for fast inference and high-throughput workloads, while maintaining strong reasoning a
- Context window
- 1,048,576
- Input modalities
- text
- Output modalities
- text
- Capabilities
- promptCaching, reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 24.4
- OpenCompass LLM: 49.1%
- OpenRouter Benchmarks: 90.109%
Inference prices
- Cached input tokens (per 1M) : $0.03 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.13 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.28 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3.5-35b-a3b-20260224
The Qwen3.5 Series 35B-A3B is a native vision-language model designed with a hybrid architecture that integrates linear attention mechanisms and a sparse mixture-of-experts model, achieving higher inference efficiency. Its overall performance is comparable to that of the Qwen3.5-
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 19.3
- OpenRouter Benchmarks: 76.333%
Inference prices
- Cached input tokens (per 1M) : $0.15 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.24 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $1.80 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3.6-27b-20260422
Qwen3.6 27B is a dense 27-billion-parameter language model from the Qwen Team at Alibaba, released in April 2026. It features hybrid multimodal capabilities — accepting text, image, and video inputs — and supports a 262,144-token context window. The model is designed for agentic
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 21.4
- OpenRouter Benchmarks: 78.98%
Inference prices
- Input tokens (per 1M) : $0.30 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $3.20 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3.5-397b-a17b-20260216
The Qwen3.5 series 397B-A17B native vision-language model is built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. It delivers state-of-the-art performance comparable to leading-e
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 21.4
- OpenCompass LLM: 88.4%
- OpenRouter Benchmarks: 79.952%
qwen/qwen3.6-35b-a3b-20260415
Qwen3.6-35B-A3B is an open-weight multimodal model from Alibaba Cloud with 35 billion total parameters and 3 billion active parameters per token. It uses a hybrid sparse mixture-of-experts architecture combining Gated DeltaNet linear attention with standard gated attention layers
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, videoInput, visionInput
Benchmarks
- Artificial Analysis: 18.2
- OpenRouter Benchmarks: 75.692%
Inference prices
- Cached input tokens (per 1M) : $0.15 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.24 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $1.80 / output_tokens_per_million · siliconFlow ↗
qwen/qwen3.5-122b-a10b-20260224
The Qwen3.5 122B-A10B native vision-language model is built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. In terms of overall performance, this model is second only to Qwen3.5-3
- Context window
- 262,144
- Input modalities
- text, image, video
- Output modalities
- text
- Capabilities
- multimodal, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 17.7
- OpenRouter Benchmarks: 78.182%
Inference prices
- Input tokens (per 1M) : $0.26 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $2.08 / output_tokens_per_million · siliconFlow ↗
deepseek/deepseek-v3.1-terminus
DeepSeek-V3.1 Terminus is an update to [DeepSeek V3.1](/deepseek/deepseek-chat-v3.1) that maintains the model's original capabilities while addressing issues reported by users, including language consistency and agent capabilities, further optimizing the model's performance in co
- Context window
- 163,840
- Input modalities
- text
- Output modalities
- text
- Capabilities
- reasoning, structuredOutputs, toolCalling
Benchmarks
- Artificial Analysis: 14.8
- LMArena: 1,417.22
- OpenRouter Benchmarks: 73.333%
Inference prices
- Input tokens (per 1M) : $0.27 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $1.00 / output_tokens_per_million · siliconFlow ↗
google/gemma-4-26b-a4b-it-20260403
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at a fraction of the compute cost. Supports multimodal input including text,
- Context window
- 262,144
- Input modalities
- image, text, video
- Output modalities
- text
- Capabilities
- multimodal, promptCaching, reasoning, structuredOutputs, toolCalling, visionInput
Benchmarks
- Artificial Analysis: 16.7
- OpenRouter Benchmarks: 65.209%
Inference prices
- Cached input tokens (per 1M) : $0.05 / cache_read_tokens_per_million · siliconFlow ↗
- Input tokens (per 1M) : $0.14 / input_tokens_per_million · siliconFlow ↗
- Output tokens (per 1M) : $0.40 / output_tokens_per_million · siliconFlow ↗