从语言到影像，从开源到 API。读懂模型能力、比较真实成本，
为你的下一个想法，找到合适的智能引擎。
文生视频：根据您输入的文本提示词+参数（可选）生成目标视频。 图生视频-首帧：根据您输入的首帧图片+文本提示词（可选）+参数（可选）生成目标视频。 官方文档：https://www.volcengine.com/docs/82379/1544106?lang=zh#02affcb8 官方计费示例： https://bytedance.larkoffice.com/wiki/FXaYwxzJ5i5Zdik32ipcWzt7nxd?table=tbl9E3YrLJ3n8Bk5&view=vewltUCWg2 注意：官方价格按 token 计费，任务完成时实际消费会进行修正，多退少补
Seedance 2.0 API（APIMart）：字节跳动文生视频与图生视频模型，生成 4–15 秒有声视频，最高 4K，支持图片、视频、音频参考与首尾帧。开通会员，API 调用立省20%起。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
Seedance 2.0 Fast API（APIMart）：功能与 Seedance 2.0 一致、生成更快，支持文生视频、图生视频与首尾帧，4–15 秒、480p/720p 输出。开通会员，API 调用立省20%起。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
Seedance 2.0 Mini API（APIMart）：功能与 Seedance 2.0 一致、提示词无字数限制，支持文生视频、图生视频与多模态参考，4–15 秒、480p/720p 输出。开通会员，API 调用立省20%起。
Seedance 2.5 标准版 多模态视频：支持最多 30 张图 + 10 段视频 + 10 段音频参考（单段 2-30s，总长 <=30s），支持 prompt+仅音频 API doc: https://gpt-best.apifox.cn/api-440251336 1080p\2k\4k 是基于 720p 原生生成后进行画质放大，有额外费用
图片编辑，OpenAI dalle 格式
文生图。OpenAI Dalle 格式
即梦4 绘图，Dalle 格式
Seedream 4.5 是字节跳动最新推出的图像多模态模型，整合了文生图、图生图、组图输出等能力，融合常识和推理能力。相比前代4.0模型生成效果大幅提升，具备更好的编辑一致性和多图融合效果，能更精准的控制画面细节，小字、小人脸生成更自然
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
This is a Vicuna-like model with only 68M parameters, which is fine-tuned from LLaMA-68m on ShareGPT data. The model is mainly developed as a base Small Speculative Model in the MCSD paper. As a comparison, it can be better aligned to the Vicuna models than LLaMA-68m with little loss of alignment to the LLaMA models.
译文排队中 · 当前显示来源原文
GGUF quantizations of dphn/Dolphin-Mistral-24B-Venice-Edition, **built from version 1.2**, with vision adapters and an importance matrix. All the Dolphin-Mistral-24B-Venice-Edition GGUFs I could find on Hugging Face were converted from an earlier release, so I built these from version 1.2, introduced in `337ce042026e`: *"Updated to version 1.2 - vision + 131k context + improved tool calling"*. Every quant was measured by KL-divergence against the BF16 weights on held-out wikitext-2 (128 chunks). `BF16` (43.92 GB) is the reference and is included for anyone wanting to re-quantize without re-downloading the safetensors.
译文排队中 · 当前显示来源原文
GGUF quantizations of Gemma-4-31B-JANG_4M-CRACK for use with llama.cpp, LM Studio, Ollama, and other GGUF-compatible inference engines. The original model uses **JANG v2 mixed-precision MLX quantization** (attention 8-bit + MLP 4-bit), which is only compatible with vMLX. Standard tools (llama.cpp, LM Studio, oMLX, mlx-lm) cannot load this format due to mixed per-layer bit widths. **Note:** Since the original was already quantized (avg 5.1 bits), the dequantized f16 intermediate is an approximation. Re-quantizing to GGUF introduces minimal additional quality loss since the attention layers were preserved at 8-bit in the original.
译文排队中 · 当前显示来源原文
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
This Dolphin is *really good* at coding, I trained with a lot of coding data. It is *very* obedient but it is not DPO tuned - so you still might need to encourage it in the system prompt as I show in the below examples. New in 2.6 - Fixed a training configuration issue that improved the quality a lot - Due to popular demand, added back samantha-based empathy data - Replaced synthia and pure-dove with Capybara This model is uncensored. I have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant to any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.
译文排队中 · 当前显示来源原文
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
My appreciation for the sponsors of Dolphin 2.8: - Crusoe Cloud - provided excellent on-demand 10xL40S node - Winston Sou - Along with a generous anonymous sponsor, donated a massive personally owned compute resource! - Abacus AI - my employer and partner in many things. This model is based on Mistral-7b-v0.2 a new base model released by MistralAI on March 23, 2024 but they have not yet published on HuggingFace. Thanks to @alpindale for converting / publishing. Dolphin is uncensored. I have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant to any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.
译文排队中 · 当前显示来源原文
Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations We have retrained our LLama-3-70b fine tune to address behavioral issues in the initial 2.9 dataset. Specifically, Systemchat was causing the model to be *too* reliant on the system prompt. Additionally, it had an occasional quirk that would cause the model to overly reference the system prompt. We also found generation length was at times not sufficient for any given task. We identified the culprit as Ultrachat. Accounting for these concerns, we removed systemchat and ultrachat from the dataset. It is otherwise identical to dolphin-2.9. Our appreciation for the sponsors of Dolphin 2.9.1: - Crusoe Cloud - provided excellent on-demand 8xL40S node - OnDemand - provided inference sponsorship
译文排队中 · 当前显示来源原文
Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations We have retrained our LLama-3-8b fine tune to address behavioral issues in the initial 2.9 dataset. Specifically, Systemchat was causing the model to be *too* reliant on the system prompt. Additionally, it had an occasional quirk that would cause the model to overly reference the system prompt. We also found generation length was at times not sufficient for any given task. We identified the culprit as Ultrachat. Accounting for these concerns, we removed systemchat and ultrachat from the dataset. It is otherwise identical to dolphin-2.9. Our appreciation for the sponsors of Dolphin 2.9.1: - Crusoe Cloud - provided excellent on-demand 8xL40S node
译文排队中 · 当前显示来源原文
可以调整厂商、分类或输入输出条件；新模型正持续收录。
调整月度 Token 用量，直观比较不同模型的预估 API 开支。
使用已采集且单位可比的 USD / 百万 Token 渠道报价。报价为采集时快照；不计缓存优惠、阶梯价、推理额外输出与税费，请以来源的当前报价为准。
先区分输入、输出和缓存价格，再使用同一任务和用量估算总费用。图像、音频和视频可能按张、时长或分辨率计费，不能直接与 Token 单价比较。
不一定。必须逐一核对具体模型版本的许可证、商用限制及使用政策。自己部署还需要考虑显存、推理资源和运维费用。
榜单取决于测试任务、版本与评测方法。建议用实际业务样本比较准确性、时延、稳定性和成本，再选择模型。
