100美元的AI音乐视频：Claude Fable 5 与 GPT-5.6 Sol 之争

我们给Claude Fable 5和GPT-5.6 Sol提供了同一首歌，一个预算、网络搜索和本地的ffmpeg，然后让它们自主指导音乐视频的制作。

我们建立了一个小型的智能代理系统，只有一个任务：给模型一首歌、一笔固定预算和一套工具，然后让出空间，让其独立制作完整音乐视频。模型会研究现有的视频模型，生成剪辑，观看自己的素材，用ffmpeg编辑，并组装出最终成片。

我们上一轮的构建对比实验的几位读者：/blog/gpt-5.6-build-off-12-models 表示他们想了解模型之间的工具使用差异，所以我们给了前沿模型一个开放式、长远任务，由每个模型自行决定研究内容、生成内容以及如何编辑。我们记录了每一次工具调用，所以你可以准确看到每个模型的工作方式（完整的记录见下文）。

我们运行了两个模型，Claude Fable 5：/models/claude-fable-5 和 GPT-5.6 Sol：/models/gpt-5.6-sol，每个模型在两种预算下运行（$25和$100），总共进行了四次实验。每次实验获取相同的歌曲（Bruno Mars和Mark Ronson的《Uptown Funk》）、一段简短文本描述以及带时间戳的歌词记录。

每个模型运行一个自主工具调用循环，使用六种工具：

一旦预算用尽，付费生成将被拒绝，但模型可以继续编辑。每条模型消息、工具调用、费用和错误都会被记录。整个系统开源在github.com/hershalb/music-video-arena：https://github.com/hershalb/music-video-arena，因此你可以自行运行。

下面的每个剪辑都是模型最终自组装的output.mp4，完整长度并包含原始歌曲。

四次实验都独立完成（没有达到步骤或时间限制），且四次都生成了有效的、完整的、包含原始歌曲的视频。

“生成花费”是计量的FAL成本，这是预算上限决定的。在$25预算下，两种模型几乎耗尽预算。$100预算下，它们分别花费了$36.57（Sol）和$48.60（Fable），因此更多预算确实产生了更多素材。这还不包括模型运行本身的成本，我们将在下文补充。

在选择自己的工具时，模型出现了分歧。四次运行中有三次是纯文本到视频。只有 GPT-5.6 Sol $25 使用了图像到视频的管线（先生成静态图像，再进行动画）。GPT-5.6 Sol $100 在一次运行中混合使用了三种不同的视频模型。

价格为 FAL 列出的费率，除非另有说明，否则按生成视频的每秒计算。Hailuo 2.3 Standard 按视频计价（每 6 秒的剪辑约 $0.28），Seedance 1.0 Pro 按 token 计价（每 5 秒 1080p 剪辑约 $0.62，上文显示为其有效的每秒费率）。每次运行生成的不同剪辑数量在 46 到 80 个之间。

每次运行的工具调用使用情况（计入尝试，包括失败的生成调用）。

每次运行的完整记录，包括每个计划、工具调用和命令，在这里： Fable 5 · $25：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/fable-25.txt, Sol · $25：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/sol-25.txt, Sol · $100：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/sol-100.txt, Fable 5 · $100：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/fable-100.txt。

“失败的调用”是返回错误的生成请求（大多为提供商的临时网络故障）。这些调用不收费，但模型在尝试重试时会消耗步骤。

预算只计算生成（FAL）的花费。加上 Claude Fable 5 的 LLM token 成本：/models/claude-fable-5（每 1M 输入/输出 $10 / $50）和 GPT-5.6 Sol：/models/gpt-5.6-sol（$5 / $30），即可得出每次运行的总成本。

对于 Claude Fable 5，仅 token 成本就为 $16.99 到 $25.05，约占每次运行总成本的 30-40%。GPT-5.6 Sol 的 token 成本尽管 token 数量相似，但保持在 $3-4 附近。

此竞技场是开源的：github.com/hershalb/music-video-arena：https://github.com/hershalb/music-video-arena。将其指向你自己的歌曲和预算，替换为你想让它们相互竞争的模型，看看它们能创造出什么。欢迎提交问题和 PR，我们非常希望收到关于此设置的反馈。

没有一个音乐视频是很出色的，但观察模型达到这一点的过程非常有趣，也确实显示了前沿级模型仍存在的明显差距。有几点需要注意：

随着模型变得越来越智能，我们拭目以待它们在更多主观/风格化任务上是否能有所提升，但就目前而言，仍然有很大的改进空间。

这里提到的每一个模型都可以通过一个账户在 TryAI 上使用，按使用付费，无需订阅。

汇聚所有前沿的 AI 模型。聊天、图像、视频和音乐——只为你使用的部分付费。

为想尝试每一个 AI 模型的人而打造。

We gave Claude Fable 5 and GPT-5.6 Sol the same song, a budget, web search, and local ffmpeg, then let each autonomously direct a music video.

We built a small agentic harness with one job: hand a model a song, a hard dollar budget, and a set of tools, then get out of the way and let it produce a full music video on its own. The model researches which video models exist, generates clips, watches its own footage, edits with ffmpeg, and assembles a final cut.

A few readers of our last build-off：/blog/gpt-5.6-build-off-12-models said they wanted to see how tool use actually varies between models, so we gave frontier-level models an open-ended, long-horizon task where each model decides on its own what to research, what to generate, and how to edit. We log every tool call, so you can see exactly how each one worked (full transcripts below).

We ran two models, Claude Fable 5：/models/claude-fable-5 and GPT-5.6 Sol：/models/gpt-5.6-sol, each at two budgets ($25 and $100), for four runs total. Every run got the same song (Bruno Mars and Mark Ronson's "Uptown Funk"), a short text description, and a time-stamped lyric transcript.

Each model ran an autonomous tool-calling loop with six tools:

Once the budget hits zero, paid generation is refused, but the model can keep editing. Every model message, tool call, charge, and error was logged. The whole harness is open source at github.com/hershalb/music-video-arena：https://github.com/hershalb/music-video-arena, so you can run it yourself.

Each clip below is the model's final, self-assembled output.mp4, full length with the original song muxed in.

All four runs finished on their own (none hit a step or time limit) and all four produced a valid, full-length video with the original song muxed in.

"Generation spend" is the metered FAL cost, which is what the budget caps. At $25 both models nearly exhausted it. At $100 they spent $36.57 (Sol) and $48.60 (Fable), so more budget did translate into more footage. It does not include the cost of running the model itself, which we add below.

Left to choose their own tools, the models diverged. Three of the four runs went pure text-to-video. Only GPT-5.6 Sol at $25 used an image-to-video pipeline (generating stills first, then animating them). GPT-5.6 Sol at $100 mixed three different video models in a single run.

Prices are FAL's listed rates, shown per second of output video unless noted. Hailuo 2.3 Standard is priced per video (about $0.28 per 6s clip), and Seedance 1.0 Pro is token-priced (~$0.62 per 5s 1080p clip, shown above as its effective per-second rate). Distinct clips generated per run ranged from 46 to 80.

How each run spent its tool calls (this counts attempts, including failed generation calls).

Each run's full transcript, every plan, tool call, and command, is here: Fable 5 · $25：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/fable-25.txt, Sol · $25：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/sol-25.txt, Sol · $100：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/sol-100.txt, Fable 5 · $100：https://d1md4c6gq9re9p.cloudfront.net/blog/music-video-arena/transcripts/fable-100.txt.

"Failed calls" are generation requests that returned an error (mostly transient network failures to the provider). They were not charged, but the model spent steps retrying them.

The budget only meters generation (FAL) spend. Adding the LLM token cost for Claude Fable 5：/models/claude-fable-5 ($10 / $50 per 1M input/output) and GPT-5.6 Sol：/models/gpt-5.6-sol ($5 / $30), gives the total cost of each run.

For Claude Fable 5, the tokens alone ran $16.99 to $25.05, about 30-40% of each run's total. GPT-5.6 Sol's token cost stayed near $3-4 despite similar token volume.

The arena is open source: github.com/hershalb/music-video-arena：https://github.com/hershalb/music-video-arena. Point it at your own song and budget, swap in whichever models you want to pit against each other, and see what they build. Issues and PRs welcome, we would love feedback on the setup.

None of the music videos were great, but watching how the models got there was pretty interesting and does show where gaps still clearly exist for frontier-level models. A few things notes:

We'll see if models can improve on more subjective/stylistic tasks as they continue to get smarter, but for now there's still a lot of room for improvement.

Every model mentioned here is available on TryAI with one account, pay-as-you-go, no subscription.

Every frontier AI model in one place. Chat, image, video, and music — pay only for what you use.

Built for people who want to try every AI model.