This is a straightforward text transformation task — no skills apply. Here's the converted markdown:
Opus 4.8: the default 1M context window is what I notice most in long sessions
Claude Code moved to Opus 4.8 a few days ago, and the default context window is now 1M. I run several long-lived automation jobs, and the old pain was predictable: once a session ran long enough, the context would fill up and I'd have to hit /compact or spin up a fresh session. With 1M as the default, that happens far less.
Opus 4.8 shipped on May 28
Anthropic released Opus 4.8 on May 28, 2026. The thing that jumped out at me was the default 1M context window. It's 1M on the Claude API, Amazon Bedrock, and Vertex AI, while Microsoft Foundry is 200k. Max output is 128k.
Pricing stays the same as 4.7. Bumping the model without raising the per-token price was a welcome detail for me.
For someone running long sessions, this was the big one
The default 1M landed harder for me than the benchmark numbers did. Before, long sessions filling up their context and breaking my flow was the most annoying part. Working with long automation-job logs attached, I hit it even more often.
$ claude
> /model
# Check that it's running as Opus 4.8 (1M context)
# Before, long sessions would fill up the context and I'd hit /compact or open a new session
# With 1M as the default, that happens noticeably less
It's only been about two days, but I'm carrying long sessions straight through more often. If you keep an entire large repo loaded while you work, you'll probably feel it even more.
The benchmarks and honesty gains are from the announcement
Per the release notes, agentic coding went from 64.3% to 69.2% and reasoning from 54.7% to 57.9%. I didn't measure these myself — these are the numbers from the announcement, so I'm just passing them along.
The honesty improvement was more interesting to me. Anthropic said Opus 4.8 lets flaws in its own code slide roughly 4x less often than 4.7. That's not something I can confirm in a few days, though, so I've left it as something to watch for a while.
Fast mode got faster and cheaper
Fast mode got some work too. On Opus 4.8 it's about 2.5x faster and costs less than before. You still turn it on with /fast, and it doesn't drop you to a smaller model — it's the same Opus running faster.
I don't usually run fast mode, so I can't say much here yet. For light tasks that need quick responses, I'm planning to leave it on and run it for a few more days.
What I'm watching over the next few days
In the end, the core of Opus 4.8 for me was the default 1M context. More than how many percentage points a benchmark moved, long sessions not breaking is the difference that adds up day after day.
Whether the honesty improvement actually shows up is something I'll only know after running the same kinds of tasks for a while. For now, just having the context fill up less is plenty to be happy about.
Sources: Anthropic — News · Claude Code — What's New
This is a first-person take after a few days of hands-on use. The benchmark and honesty figures are copied straight from Anthropic's announcement, and this post is not sponsored in any form.
Original with full infographics and visual structure: https://jessinvestment.com/opus-4-8s-default-1m-context-window-is-the-upgrade-i-notice-most/
Top comments (0)