Is Claude Code truly faster with message threads? I did the math on my logs

ZDNET’s crucial takeaways

• Anthropic’s message threads make Claude Code a little more responsive.
• My tests recommend cost savings of 2 to 4 minutes per coding session.
• You get faster reactions, however no cost savings on token expenses.

My real-world speed quotes

This function is on by default for many Claude Code users. API users can turn it on utilizing the “thread” criterion for API demands.

Token usage stays the same, so this ability does not offer an expense savings. The 35% cost savings is Anthropic’s measurement.

: Claude Code’s revised jobs includes AI orchestration, however regional designers need to wait

Timing information from approximately 980 turns in my own Claude Code logs before this function appeared programs that Claude generally started an action about 3 seconds after each mid-task demand when running contexts of 500,000 tokens and up. Throughout 5 hectic nights last month, my discussions made approximately 100 to 225 design hires a 4 hour stretch.

Utilizing Anthropic’s reported 35% typical time-to-first token decrease in a fast forecast, Claude approximated a 2- to 4-minute cost savings per 4 hour work session.

Message threads minimize duplicated uploads

Representatives frequently send out a lot of demands to the design throughout each discussion. Before message threads, each message sent out the whole discussion with each demand.

Picture if you’re in a long texting chat with a pal. Now think of if whenever you sent out a text to that good friend, you likewise needed to send out every other text and every other reply together with that initial text. That’s how model messaging has actually constantly worked. Other than now, with message threads, there’s a much better method.

: 75% of designers I surveyed choose Claude Code– here’s why they pick it over Codex

With message threads switched on, representatives just need to send what’s altered considering that the last turn. This lowers the quantity of information transmission and increases responsiveness, what Anthropic calls time-to-first-token.

The advantage is seen most highly in sessions that have a context of 500,000 to a 1 million tokens. About half of my sessions strike that context ceiling several times a night.

Just how much quicker is it?

Anthropic reports that Claude’s message threads lower typical time-to-first-token by 35% for big demands. While my approximated cost savings of a number of minutes per session isn’t big, the advantage increases with extremely long-running coding representatives, and representatives that transfer files and image-heavy work.

For computer-using representatives, all those screenshots “stick with the thread rather of being re-uploaded on every turn,” according to the Anthropic post on the function.

: How I made a paid Mac app in 11 days with Claude– without composing a single line of code

While the time cost savings might not appear substantial, in bigger business implementations or releases where you desire the AI to react much faster, any decrease in latency is a huge offer and will be seen in any application of Claude innovation that has a great deal of back-and-forth turns with big information payloads.

Faster does not indicate less expensive, or more costly

Despite the fact that the addition of threads to the Messages API can lower latency, token use and for that reason expense will not decrease. That’s since the design itself will still process the complete discussion throughout every turn. Rate limitations likewise do not alter.

Message threads and timely caching can interact to lower work. Trigger caching has actually been around for about 2 years now, and it does conserve cash. That’s due to the fact that timely caching enables the design to prevent re-processing triggers over and over (for instance, in looping analysis calls).

On the other hand, the message threads ability does use an additional smidge of efficiency, at no additional expense. There’s that.

It likewise will not make Claude smarter

Context is the size of the complete session that Claude is presently dealing with. The rough human equivalent is present working or short-term memory. The majority of Claude Code sessions max out at a million token context, where a locally-hosted AI on my 128GB Mac Studio manages a context no larger than 64K, or about 6% of the Claude Code context size.

: After the vibe-coding rush comes the debugging hangover

Compaction is a procedure where the AI tries to decrease and get rid of additional info to fit the existing session’s develop into the offered context. An oversimplified analog is when you conserve a JPEG with quality turned method down in order to conserve area. You lose a great deal of the visual fidelity, however you still get the essence.

The bigger the context you have, the longer you can precede the session needs to start compaction. There’s a gotcha. The bigger the context you have, the more possibility there is the design will misplace what it’s dealing with or make thinking mistakes.

Specific one-off demands likewise will not see an enhancement from message threads. Any discussion that postpones longer than 5 minutes in between turns will send out the complete message contents and not simply the brand-new things.

This function does not get rid of context management or compaction. If your context is rather big, it will make actions a bit snappier.

A little however valuable speed increase

In general, message threads are an advantage, a strong if little development that increases responsiveness. While the function does not minimize expenses, there’s likewise no extra expense for the extra efficiency, which is itself a win.

You can follow my daily task updates on social networks. Make certain to register for my weekly upgrade newsletterand follow me on Twitter/X at @DavidGewirtzon Facebook at Facebook.com/ DavidGewirtzon Instagram at Instagram.com/ DavidGewirtzon Bluesky at @DavidGewirtz. comand on YouTube at YouTube.com/ DavidGewirtzTV