Opus 4.8 Launched, but Fast Mode Was the Real Win for Me

#claudecode #opus48 #fastmode #aicoding

Opus 4.8 Launched, but Fast Mode Was the Real Win for Me

Yesterday (May 28), Anthropic rolled Opus 4.8 out globally. The announcement was about coding scores climbing again, but since I run Claude Code all day long, something else caught my eye before the benchmark table did: the line saying fast mode got 2.5x faster and 3x cheaper.

01. Same price, just a better model

Opus 4.8 went out to everyone the moment it was announced yesterday. According to Anthropic, the core of it is doing agentic work — like coding — better.

The good news is the price. Anthropic says the per-token rate stays the same between Opus 4.7 and 4.8. It's one of those cases where the model quietly levels up but the bill doesn't change, so for an existing user there was no added burden to worry about.

02. How the benchmark table changed

These are the 4.7-vs-4.8 numbers Anthropic published. Let me note up front that I didn't measure these — they're straight from the announcement notes.

Metric	4.7	4.8
Agentic coding	64.3%	69.2%
Tool-use reasoning	54.7%	57.9%
Knowledge-work score	1753	1890
Computer use	82.8%	83.4%

They also say high effort is now the default for coding tasks. The explanation is that you spend roughly the same tokens as 4.7 but get better results.

03. For me, fast mode was the real story

What actually touched my day-to-day, more than the scores, was the fast mode side. Per the announcement, the fast mode variant now runs 2.5x faster than before while costing 3x less.

I run several Claude Code jobs at once — AI ecosystem monitoring, stock market monitoring, writing posts. In a setup like that, faster responses and cheaper tokens hit you immediately. Here's how I usually launch it.

$ claude --model opus      # this is how I usually launch it
> /fast                     # toggle fast mode (Opus 4.8/4.7/4.6 supported)

# I run a bunch of jobs in parallel, so
# faster, cheaper responses pile straight up into cost · time savings

On top of that, Claude Code's rate limits went up too. With the extra · max settings you can funnel more tokens through. Heavy work that kept hitting the ceiling should breathe a little easier now.

04. I can't speak to the coding feel yet

Honestly, I can't promise "coding got 5pp better" just from the benchmark table. The model dropped only yesterday, so it'll take a few more days of use before I know what's actually different from 4.7 in my own work.

What I haven't verified yet — The only thing I measured myself is the fast mode speed; I haven't done a hands-on comparison of coding accuracy. The benchmark table is copied straight from the announcement notes, so I'll write up the real-world feel separately.

Still, getting frozen pricing + faster fast mode + higher limits all in one drop was a pretty big bundle for someone running automation. On that part alone, I was glad from day one of the announcement.

Eddie's take — Honestly, I tend to stay pretty unmoved every time a model's benchmark table goes up. You only learn the real difference after a few days of use anyway. But this time, making fast mode faster and cheaper without raising the price was a far more concrete change — for someone running it every day — than a few percentage points on a score.

Sources

This post summarizes an external announcement. The score and price figures are based on Anthropic's announcement notes, and aside from fast mode speed, I did not personally measure the coding experience. No sponsorship of any kind was received.

Canonical: https://jessinvestment.com/opus-4-8-launched-but-fast-mode-was-the-real-win-for-me/

Original with full infographics and visual structure: https://jessinvestment.com/opus-4-8-launched-but-fast-mode-was-the-real-win-for-me/