完全免费的Opus-4.8 CODER：这真的非常有用！

translation:

YouTube Transcript

语言：英语（自动生成）（en）

[0:02] [音乐]

[0:05] >> HI.Hello to another video. So, Claude

[0:08] Opus 4.8 is out now, and in my testing,

[0:11] this is(probability the strongest coding

[0:12] model available right now. But, the

[0:14] problem is obvious. Opus is expensive.

[0:17] If you use it directly through the API,

[0:20] it costs $5 per million dollar input

[0:22] tokens and $25 per million output

[0:24] tokens. And if you're using it for real

[0:27] coding work, especially with large

[0:29] repositories, long context, plan,

[0:32] tool calls, and verification, those

[0:34] tokens can disappear very fast. So, in

[0:36] this video, I want to show you a couple

[0:38] of places where you can try Opus 4.8

[0:40] without paying up front. And the first

[0:42] one is Verdant. Now,

[0:44] I have already covered Verdant multiple

[0:46] times on the channel, and I still think

[0:47] it is one of the most interesting AI

[0:49] coding tools right now, because it is

[0:51] not just another chat box inside vs

[0:53] Code. Verdant is more like an agoric

[0:55] coding workspace.

[0:56] The important part for this video is

[0:58] that Verdant currently gives new users a

[1:00] 7-day free trial with free credits. And

[1:02] from what I have seen, you do not need

[1:04] to put in a card first. You just

[1:06] download it, sign in, and it works. So,

[1:08] this is the easiest place I would start

[1:10] if you just want to test a powerful

[1:11] coding agent without doing the whole API

[1:13] key setup, billing setup, and token

[1:15] anxiety thing. Just go to Verdant,

[1:17] download the desktopapp, or use the V**

[1:19] Code extension, sign in, and you should

[1:20] get the free trial credits. Then, inside

[1:23] the model selector, choose Claude Opus

[1:25] 4.8 if it is available on your account.

[1:27] If it is still rolling out, just update

[1:29] theapp or check again, because these

[1:31] model launches usually appear in stages.

[1:33] And the reason I like Verdant for this

[1:34] is not only because of the free trial,

[1:36] it is because the workflow around the

[1:37] model is actually good. Like I said in

[1:39] my previous Verdant videos, Verdant

[1:41] feels less like aծ coding toy and

[1:43] more like a professional multi. So,

[1:45] instead of giving one large task to one

[1:46] agent and waiting forever, you can split

[1:49] the work.

[1:50] One agent can work on the back end,

[1:52] another one can work on the front end,

[1:53] another one can write tests, and another

[1:55] one can review the changes. And these

[1:57] agents can work in isolatedgit work

[1:59] trees, which is very important, because

[2:01] if you have used coding agents before,

[2:03] you know how quickly things can go wrong

[2:05] when two agents edit the same files or

[2:07] when an agent destroys something that

[2:08] was already working. With Verdant, each

[2:10] task can run in its own working space,

[2:12] almost like its own branch. So, the main

[2:14] repo does not get cooked immediately.

[2:16] You can inspect the diff, review the

[2:17] changes, and then merge it back when it

[2:19] actually looks good. That is one of the

[2:21] largest reasons I like Verdant. It gives

[2:23] you the feeling of managing dev team,

[2:26] but without every agent step on each

[2:27] other. Another feature I real like is

[2:29] the plan-first workflow. You can switch

[2:32] to plan mode, describe what you want,

[2:34] and Verdant does not just start editing

[2:35] files instantly. It discuss the

[2:37] approach, asks clarifying questions, and

[2:40] then gives you a plan. Once you approach it, then it starts making changes. This is especially useful with Opus 4.8, because Opus 4.8 is very good at

[2:44] long-horizon tasks and agoric work, but even a good model still needs structure. If you just tell it build my whole SaaS, it might do something, but if you make it plan first, break the task into steps, and verify each part, then you get much better results. Verdant also has a good diff review experience. You can see exactly what changed, inspect the files, and decide what to accept. This matters a lot, because with these front coding models, the output can look very convex even when something subtle is wrong. So, I would not use the free trial by just asking random questions. If you want to actually test Opus 4.8 properly, give Verdant a real coding task.

[3:23] For example, open one of your existing projects and ask it to implement a feature you have been delaying.

[3:25] asked it to refine a messy component.

[3:27] asked it to add tests. asked it to fix a real bug. asked it to make a ui more polished. That is where you will actually feel whether Opus 4.8 is worth it.

[3:41] because Opus 4.8 is not interesting because it can write a to-doapp. Every model can write a to-doapp now. It is interesting because it is better at staying on task for longer, using tools more efficiently, catching its own mistakes, and working through complex code changes. And verdant is a good wrapper around that because it gives you the agoric worked flow, isolated])])])])))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))

为了将技术文章在中文和英文之间进行高质量翻译，我们需要遵循以下步骤：

1. 保持 Markdown 格式不变：确保标题、列表、代码块、链接等保持原样。

1. 技术术语保持准确一致： common术语保留英文（如 API、API、Docker 等）。

1. 翻译要自然流畅：确保翻译自然、流畅，不要逐字翻译。

1. 代码块内容不翻译：保持代码块内容不变。

1. 图片链接和 URL 保持原样：保持图片链接和 URL 不变。

现在，我们来翻译给定的 Markdown文：

[4:00] workspaces, planning, review, and

[4:02] multitasking. So, that is the first

[4:04] option. You can basic

[4:06] Verident, use the 7-day free trial,

[4:09] and test Opus 4.8 there without paying

[4:11] up front. Now, the second place is Kiro.

[4:14] Kiro has also added Claude Opus 4.8, and

[4:17] this one is also very interesting

[4:18] because Kiro is giving people a way to

[4:20] try their paid plan without paying up

[4:21] front. Their power plan is normally $200

[4:24] per month, and it gives you 10,000

[4:26] credits per month. That is the plan you

[4:27] would use if you want to serious test

[4:29] the high-end models like Opus. And right

[4:31] now, the checkout can show zero due

[4:32] today for the trial. So, basically, you

[4:35] get a chance to try Kiro power and Opus

[4:36] 4.8 before paying the full $200.

[4:39] That is quite a big deal because

[4:40] normally when people hear $200 plan,

[4:43] they straight away ignore it. And I get

[4:45] that. $200 per month is a lot of money

[4:47] for most people, but if the trial lets

[4:49] you test it first, then at least you can

[4:51] see whether the workflow is actually

[4:52] useful for your projects before

[4:54] commit. In Kiro, Opus 4.8 is

[4:57] available in the IDE,_CLI, and web. It

[5:00] comes with a 1 million token context

[5:01] window, a 128k max output limit, and it

[5:05] has the same 2.2x credit multiplier as

[5:07] Opus 4.7. That means it is not a cheap

[5:10] model inside Kiro's credit system,

[5:12] but it is very powerful.

[5:14] So, if you are testing it, do not waste

[5:16] the credits on small prompts like explain

[5:19] this function.

[5:20] Use Sonnet or cheaper models for that.

[5:22] Use Opus 4.8 for the hard stuff. For

[5:25] example, use it when you want to

[5:27] understand a large codebase, plan a

[5:29] complex migration, fix a bug that spans

[5:31] multiple files, build a feature from a

[5:33] spec, review a difficult PR, or do

[5:36] something that needs long context and

[5:37] careful reasoning.

[5:39] That is where Opus actually makes sense.

[5:41] And this is also where Kiro's

[5:42] spec-driven workflow can be useful. You

[5:44] can describe the feature, let it build a

[5:46] spec, refine that.spec, and then make it

[5:48] implement the change. That is much

[5:50] better than just throwing a

[5:51]ovies seed at the model and希望 it figures

[5:53] everything out. I would also

[5:54] using Kiro_CLI if you like terminal

[5:56] workups. With the latest Kiro_CLI

[5:59] version, you can select Opus 4.8 from

[6:01] the model selector and use it for

[6:03] agubic coding tasks directly from the

[6:05] command line. So, if you prefer not to

[6:07] live inside an IDE, that is a good

[6:09] option as well. Now, a quickly warning

[6:11] here. These free trials are great, but

[6:14] do not forget that they are still trials for paid

[6:17] products. So, check the billing screen,

[6:19] check the renewal date, and cancel

[6:22] before the trial ends if you do not want

[6:23] to pay.

[6:24] I know this looks obvious, but people

[6:26] forget this and then blame the tool

[6:28] later. Also, these offers can change by

[6:30] the time you watch this. Verdant might

[6:32] change the free trial credits. Kiro

[6:34] might change the check out flow, or Opus

[6:36] 4.8 might have different availability

[6:38] depending on region or account type. So,

[6:42] always check the pricing page before

[6:43] sign up. But, as of now, these are

[6:45] two very good ways to try Opus 4.8

[6:47] without paying up front. Verdant is the

[6:49] one I would

[6:51] first if you want a full agubic

[6:53] work space with parallel work trees, planning, and a

[6:55] real natural flow. Kiro is the one I would

[6:57] if you want to test Opus 4.8 inside Kiro's

[6:59] IDE,_CLI, and web worked flow, especially with the power plan

[7:02] trial and the large credit pool.

[7:07] And honestly, this is the best way to

[7:09] evaluate these models.

[7:10] Do not just listen to Benchmarks.

[7:12] Do not just listen to people saying this

[7:14] model is cracked or this model is brown.

[7:17] Open your own project, give it a real

[7:19] task, and see what happens. asks it to

[7:21] build something useful. asks it to fix

[7:23] something annoying. asks it to review

[7:25] something you actually care about. That

[7:27] is how you will know whether Opus 4.8 is

[7:29] worth using for your work flow. For me,

[7:32] Opus 4.8 is genuine very power.

[7:35] It is expensive, but if you can access

[7:36] it through these trials, then you should

[7:38] absolute test it. So, start with

[7:40] Verdant because the 7-day trial is very

**[7