Meta 发布 Muse Code 与 Muse Spark 1.2，聚焦长序列智能体工具调用

介绍 Muse Code 和 Muse Spark 1.2: https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2（来源：https://news.ycombinator.com/item?id=49187575） 这是又一个证据表明，如今任何模型最重要的特性是长序列的自主工具调用。Meta 推出了他们自己的编码代理，以实现这一点！

Muse Spark 1.2 是针对编码的 Muse Spark 1.1 更新，改进了代码生成、复杂调试、代码库理解以及端到端开发者工作流。在 Muse Spark 1.2 中，我们显著增加了编码任务的训练计算量，同时扩展了训练环境的多样性。该模型在其他关键领域，如通用代理，也保持了强大能力。[...]

我们将 Muse Spark 1.2 与 Muse Code 共同训练，以确保模型在配合使用时表现出最佳性能和编码可用性。训练包括拒绝采样的 harness 轨迹和针对目标、压缩及子代理的配方优化，以及 Muse Code 工具集的整合，以最大化 harness 兼容性。[...]

Muse Spark 1.2 在长远程编码任务上进行了广泛训练，包括整个代码库生成、大型端到端项目以及自动研究。

这是由 Muse Spark 1.2 生成的一只骑自行车的鹈鹕 SVG：https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Fce974a21202b0595e36ec2a5ddb51480:

你可以在这里看到 7 月 9 日的 Spark 1.1 鹈鹕：https://simonwillison.net/2026/Jul/9/muse-spark-1-1/。我认为 1.2 版本的鹈鹕有一个小但实质性的改进。

关于定价的一个有趣变化是，该模型提供了两个不同的模型 ID：https://developer.meta.com/ai/models/muse-spark/。muse-spark-1.2 的价格为每百万输入 1.25 美元、每百万输出 4.25 美元——接近 Gemini 3.6 Flash（1.50/7.50 美元）——但如果你同意让 Meta 使用你的数据“来改进我们的产品”，你可以使用 muse-spark-1.2-contributor，价格为 0.10/0.20 美元——大幅折扣，更接近 GPT-5.6 Luna（0.20/1.20 美元）和 Gemini 3.1 Flash-Lite（0.25/1.50 美元）。

我已将这些新价格添加到 llm-prices.com：https://www.llm-prices.com/#sel=muse-spark-1.2%2Cmuse-spark-1.2-contributor。

这是 Simon Willison 在 2026 年 8 月 5 日发布的链接文章：/2026/Aug/5/。

每月赞助我 10 美元，即可获得本月最重要的 LLM 发展精选电子邮件摘要。

Introducing Muse Code and Muse Spark 1.2：https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2 (via：https://news.ycombinator.com/item?id=49187575) Yet more evidence that the most important characteristic of any model these days is long-sequence agentic tool calling. Meta shipped their own coding agent as part of getting that to work!

Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. In Muse Spark 1.2, we significantly scaled up training compute on coding tasks while expanding training environment diversity. The model also maintains its strength in other key areas like general agents. [...]

We co-trained Muse Spark 1.2 with Muse Code to ensure the model exhibits its best performance and coding usability when paired together. The training included rejection sampled harness trajectories and recipe optimizations for goals, compaction, and subagents, alongside the integration of the Muse Code toolset to maximize harness compatibility. [...]

Muse Spark 1.2 was extensively trained on long-horizon coding tasks, including whole-repository generation, large end-to-end projects, and auto-research.

Here's a pelican riding a bicycle SVG produced by Muse Spark 1.2：https://tools.simonwillison.net/markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Fce974a21202b0595e36ec2a5ddb51480:

You can see the Spark 1.1 pelican from 9th July here：https://simonwillison.net/2026/Jul/9/muse-spark-1-1/. I think the 1.2 pelican is a small but material improvement.

An interesting twist on pricing is that the model is offered：https://developer.meta.com/ai/models/muse-spark/ as two different model IDs. muse-spark-1.2 is priced at $1.25/million input and $4.25/million output - close to Gemini 3.6 Flash ($1.50/$7.50) - but if you agree to let Meta use your data "to improve our products" you can use muse-spark-1.2-contributor which is $0.10/$0.20 - a huge discount, closer to GPT-5.6 Luna ($0.20/$1.20) and Gemini 3.1 Flash-Lite ($0.25/$1.50).

I added those new prices to llm-prices.com：https://www.llm-prices.com/#sel=muse-spark-1.2%2Cmuse-spark-1.2-contributor.

This is a link post by Simon Willison, posted on 5th August 2026：/2026/Aug/5/.

Sponsor me for $10/month and get a curated email digest of the month's most important LLM developments.