Claude Code Just Got 3X FASTER (And It's Free) — Transcript

Full transcript

1. 0:00Claude Code just got about three times
2. 0:02faster, and you don't pay a penny extra
3. 0:04to get it. Anthropic says the wait
4. 0:06before Claude starts answering dropped
5. 0:08from about 4 seconds to about 1 and 1/3.
6. 0:11Claude sessions now warm up while you're
7. 0:13still typing. There's a catch. That
8. 0:15three times number isn't what most
9. 0:17people think it means. I'll show you
10. 0:19what really changed, how to get it, and
11. 0:21the one speed mistake that costs you
12. 0:23money. Hey there. So, if you're new
13. 0:25here, I'm Nick Pontis' AI avatar. While
14. 0:28the real Nick is busy helping businesses
15. 0:30with minor marketing, Hawaii's fastest
16. 0:32growing marketing agency, I'm here
17. 0:34[snorts] dropping the latest AI hacks,
18. 0:36tools, and money-making strategies. The
19. 0:38real Nick reads every single comment on
20. 0:41these videos, so make sure you comment
21. 0:42below. And hey, if you're serious about
22. 0:45landing some AI subscription-based
23. 0:46customers, grab my AI cash flow
24. 0:48masterclass that I am currently offering
25. 0:50for free. The link's in the description.
26. 0:52All right. Now, let's get into it.
27. 0:56Let's start with what's actually been
28. 0:57said. Around October 16th, Anthropic's
29. 1:00Claude Devs account posted about Claude
30. 1:02Code speed. Two new sites then wrote it
31. 1:05up. They quote the same numbers. Here
32. 1:07they are. For Claude Code running on
33. 1:09your own computer, the wait for the
34. 1:11first reply used to be about 4 seconds
35. 1:13on the slow end. Now, it's about 1 and
36. 1:151/3 seconds. That's nearly three times
37. 1:17faster. What does the slow end mean?
38. 1:20Think of 10 replies. Nine of them now
39. 1:23start in about 1 and 1/3 seconds or
40. 1:25less.
41. 1:26Why did it get faster? Claude Code sends
42. 1:28your whole conversation to the model
43. 1:30every time you hit enter. Anthropic now
44. 1:32squeezes that package before it goes
45. 1:34out. The tool for that's called Gzip.
46. 1:36It's like zipping a folder before you
47. 1:38email it. A smaller package travels
48. 1:40faster. Anthropic's own release notes
49. 1:42list this. They say large requests now
50. 1:45use that compression. I checked that
51. 1:46page myself. The same reports say other
52. 1:49Anthropic products got a smaller boost.
53. 1:51Claude Chat, Cowork, and the API are
54. 1:54about 1 and 1/2 times faster. Here's
55. 1:56something to remember. a faster start
56. 1:58doesn't make the answer smarter. Judge
57. 2:00the work, not the stopwatch.
58. 2:05So, why do a few seconds matter? Because
59. 2:07of how you work. You don't send one
60. 2:09message, you send dozens. Every single
61. 2:12message has a small weight at the start,
62. 2:14so those weights add up. One outside
63. 2:16test shows how this adds up. A team
64. 2:18called Agents Room looked at about
65. 2:2020,000 turns from their own Claude code
66. 2:22sessions. They found every turn pays a
67. 2:24fixed cost before the first word shows
68. 2:26up. They put it at roughly 1 second. So,
69. 2:29a job with many tiny steps pays that
70. 2:32cost again and again. A quick answer
71. 2:34feels slower per word for the same
72. 2:36reason. The fixed weight is a bigger
73. 2:39slice of a short reply.
74. 2:40That test ran before this change on one
75. 2:43person's computer. So, treat it as a way
76. 2:45to think, not as proof of the new speed.
77. 2:50Claude sessions are Claude code running
78. 2:52on Anthropic's computers, not yours. The
79. 2:54reports say the wait there dropped from
80. 2:56about 17 seconds to about six. That's
81. 2:59roughly 2.8 times faster. How? The
82. 3:02Claude session now starts getting ready
83. 3:03as soon as you begin typing. It doesn't
84. 3:05wait for you to hit enter. One report
85. 3:07adds a fair warning. The startup work
86. 3:09isn't gone, it's just hidden behind your
87. 3:11typing. And some sessions get warmed up
88. 3:13and never get used.
89. 3:17Now, the fine print. This is the wait
90. 3:19before the first word. It isn't how fast
91. 3:21Claude writes after that. A long answer
92. 3:23still takes time to finish.
93. 3:25These numbers come from Anthropic's own
94. 3:27measurements as reported by two news
95. 3:29sites. I couldn't find the original post
96. 3:31on a page I trust to quote. The release
97. 3:33notes confirm the compression. The
98. 3:35Claude typing change shows up only in
99. 3:37the news reports.
100. 3:38And I haven't timed any of it myself.
101. 3:40Treat about three times as a rough
102. 3:42guide, not a promise for every session.
103. 3:45Your results will vary. Reports say the
104. 3:47gain is biggest in long conversations,
105. 3:49since those carry the most to compress.
106. 3:51Your internet speed matters, too. Try
107. 3:53this. Time your next three replies with
108. 3:55a stopwatch, then update and time three
109. 3:58more. Your own numbers beat any
110. 4:00headline.
111. 4:03This didn't happen fast for me. It took
112. 4:05years, a lot of trial and error before
113. 4:08anything really worked. So, when a tool
114. 4:10gets three times faster, I'm happy.
115. 4:13But, I know speed isn't a shortcut to
116. 4:14skill. A fast tool just lets you try
117. 4:17more things in one afternoon. And more
118. 4:19tries is how I finally learned. Real
119. 4:21quick, before I keep going. If you've
120. 4:23been watching this and thinking, "Okay,
121. 4:25this AI stuff is interesting, but how do
122. 4:28I actually turn it into income?" That's
123. 4:29exactly why I built the free AI cash
124. 4:32flow master class. This isn't some
125. 4:33surface level overview. It's a real
126. 4:36walk-through of how people are using
127. 4:37simple AI workflows right now to build
128. 4:39actual income streams online. Inside,
129. 4:42you'll see the specific AI services
130. 4:44businesses are already paying for.
131. 4:46Workflows you can set up without any
132. 4:48coding or technical background. How to
133. 4:50spot the businesses already dealing with
134. 4:52what I call AI guilt, meaning they know
135. 4:54they're behind, they want help, and
136. 4:56they're ready to pay for it. And how to
137. 4:58turn that into income that keeps coming
138. 4:59back. In other words, this is the bridge
139. 5:01between watching a video like this one
140. 5:03and actually getting paid. If you want
141. 5:05to see exactly how it works, the link's
142. 5:07in the description and pinned in the
143. 5:09comments. Now, let's get back into it.
144. 5:14Good news. The first part costs nothing.
145. 5:16Step one, update Claude code. In the
146. 5:19terminal, type Claude update and press
147. 5:21enter. The release notes list the
148. 5:23compression change in a recent version,
149. 5:25so you want the newest one. Step two,
150. 5:28close Claude code and open it again, so
151. 5:30the new version is the one running. Step
152. 5:32three, start a normal session and send a
153. 5:35short message. You don't turn anything
154. 5:37on. The compression is automatic.
155. 5:39Anthropic's notes say you can switch it
156. 5:40off with a setting, but there's no
157. 5:42reason to. Step four, if you use cloud
158. 5:45sessions, open one and start typing. The
159. 5:48reports say the warm-up starts by
160. 5:49itself. Step five.
161. 5:51Use that stopwatch test from earlier,
162. 5:54write down the before and after. If you
163. 5:56don't see a change, don't panic. A short
164. 5:59chat might feel the same. Try a long one
165. 6:01before you decide. One more thing, if
166. 6:03you use Claude code through a company
167. 6:05account or a different provider, ask
168. 6:07whoever runs it whether the update has
169. 6:09reached you.
170. 6:12Now, how could this help your client
171. 6:13work? I haven't tested these. They're
172. 6:15ideas to try. Idea one. Fix a page live
173. 6:19on a call. Tell Claude to change the
174. 6:21headline, the button color, or the phone
175. 6:23number while your client watches. Quick
176. 6:25replies make that feel smooth.
177. 6:27Idea two. Run many tiny edits. A small
178. 6:31business site often needs 20 little
179. 6:33changes. Fast starts help most when
180. 6:35there are lots of small steps.
181. 6:37Idea three. Batch the boring stuff. Ask
182. 6:40it to clean a messy customer list or
183. 6:42draft follow-up messages for you to
184. 6:44check and send. Notice the pattern.
185. 6:46These are small jobs with lots of back
186. 6:48and forth. That's where a quicker reply
187. 6:50helps the most. Quick warning, some
188. 6:52speed is free, some speed costs extra.
189. 6:55Know which one you're using before you
190. 6:56click.
191. 6:59There's a second kind of speed, and it
192. 7:01isn't free. It's called fast mode.
193. 7:03Anthropic's docs say fast mode makes
194. 7:05Opus, one of their models, up to two and
195. 7:08a half times faster. Same model, same
196. 7:10quality.
197. 7:11But you pay a higher price for every
198. 7:13word it handles.
199. 7:15It's a different thing from the update.
200. 7:16The update is about the wait before the
201. 7:18first word. Fast mode is about how
202. 7:20quickly the reply gets written. You turn
203. 7:22it on by typing {slash} fast, then
204. 7:25pressing space and enter. Type it again
205. 7:28to turn it off. Here are the catches
206. 7:29from the docs. It works on certain Opus
207. 7:32models, not on Sonnet or Haiku. On a Pro
208. 7:35or Max plan, it's paid for with extra
209. 7:37usage credits, and you have to turn
210. 7:39those on first. It isn't covered by your
211. 7:41normal plan limits. On team plans, an
212. 7:44owner has to allow it first. One price
213. 7:46in the docs is $8 per million tokens in
214. 7:49and $40 out for one Opus model. Tokens
215. 7:52are small chunks of words. Another model
216. 7:54costs more, and it's still a research
217. 7:56preview, so it may change. Fast mode
218. 7:59also works in cloud sessions. Type
219. 8:01{slash} fast on inside the session. It
220. 8:04stays on for that session only. A small
221. 8:06arrow icon shows next to the prompt when
222. 8:08it's on. Also, turn it on at the start
223. 8:10of a session. If you switch on mid-chat,
224. 8:13you pay the full price for everything
225. 8:14already said once. If you run out of
226. 8:17usage credits mid-session, Claude code
227. 8:19drops back to normal speed and price, so
228. 8:21your work keeps going. When is it worth
229. 8:23it? The docs say quick back and forth,
230. 8:25like fixing a page live, not long jobs
231. 8:28that run on their own. There's also a
232. 8:30cheaper way to go faster, lower the
233. 8:32effort level. That makes Claude think
234. 8:34less before it answers. The docs say it
235. 8:36can be faster, but it may be less
236. 8:38careful on hard tasks. You can combine
237. 8:40it with fast mode on simple jobs. Just
238. 8:43don't use it on work where mistakes cost
239. 8:44you. Pick one real client task this
240. 8:47week. Run it start to finish. One real
241. 8:50job teaches you more than 10 headlines.
242. 8:54Here's the catch. A faster Claude
243. 8:56doesn't land you a client. Someone still
244. 8:58has to trust you and pay you. Speed only
245. 9:01saves you time on the work after that.
246. 9:03Faster tools also mean you can try more
247. 9:05things in one afternoon, and that's the
248. 9:07real win. You also still have to check
249. 9:09what it makes. A fast wrong answer is
250. 9:11still wrong. Read the work before you
251. 9:13send it to a client. Everything here
252. 9:15comes from public pages I opened myself.
253. 9:17If Anthropic shares more details, the
254. 9:19numbers could shift. So, is the title
255. 9:22true? Mostly. The first reply wait did
256. 9:25drop by about three times on local
257. 9:27sessions, according to Anthropic's
258. 9:29numbers. But, that's the wait, not the
259. 9:31whole job, and nobody can promise you
260. 9:33results from it. Here's how I'd use it
261. 9:35this week. Update time three replies.
262. 9:37Try one small client task. Decide if
263. 9:40fast mode is worth the cost for that
264. 9:42job. And remember, I can't promise any
265. 9:44income from this. Start with one useful
266. 9:47result for one person. Before you go,
267. 9:49let's bring this back to you. If
268. 9:51everything I just walked through has you
269. 9:52thinking about how to actually make
270. 9:54money with AI, not just play around with
271. 9:56it, that's exactly what the free AI cash
272. 9:58flow masterclass is built for. It's not
273. 10:00theory. It's a real step-by-step
274. 10:03breakdown of how people are using simple
275. 10:05AI workflows to build income streams
276. 10:07right now. Inside, you'll get access to
277. 10:09a free 30-day trial of the exact
278. 10:11software I use to land clients and run
279. 10:13my own business, on top of everything
280. 10:15else in the training. You'll also get
281. 10:17access to the AI control room, my
282. 10:19private community of AI operators and
283. 10:21business owners building right alongside
284. 10:22you. If you want in, the link's in the
285. 10:24description and pinned in the top
286. 10:26comment. Go check it out, then come back
287. 10:29and let me know what you build with it.