Anthropic's newest Opus has a million-token memory. The quiet launch is the tell.
Jump to a section
See the story at a glance
If you use Claude for anything — writing, research, a work task, coding — something changed Thursday that's worth knowing about.
Claude Code v2.1.219 shipped July 24 with a single changelog line: "Added Claude Opus 5 (claude-opus-5), now the default Opus model — 1M context, fast mode at $10/$50 per Mtok." No separate blog post. No benchmark table. No head-to-head against GPT-5.6. Just the changelog entry, and a new default.
That's interesting for what it isn't. I want to sit with that before getting into what it is.
Here's the object lesson on that number. A model's context window is how much it can hold in its head at once. Think of it like the size of a desk. Bigger desk, more documents spread out at the same time. One million tokens is roughly 750,000 words. Three full-length novels. A year's worth of a professional's email archive. A large codebase laid flat. Opus 5 can read all of that in a single conversation and reason across it. Opus 4.7 couldn't.
What that means for everyday Claude users: longer conversations without the model losing track of what you said three hours ago. For builders: full-codebase analysis, large document synthesis, or extended agent sessions without constant context management.
Source spread
- Claude Code v2.1.219 release notes — builder. Primary source: the model announcement, context window, pricing, and the default-Opus swap.
- Anthropic Python SDK v0.120.0 changelog — builder. Adds
claude-opus-5to the model list, plus tool addition/removal blocks andtool_changeevents shipped alongside. - Anthropic news page — no official announcement page exists at time of writing. That absence is part of the analysis.
What's real
The 1M context is real and verified. And it's not new to the Opus tier — Opus 4.8 launched in May with 1M context as the default. Opus 5 carries it forward. This is table stakes now, not a headline feature.
The $10/$50 fast-mode pricing is real. That's $10 per million input tokens and $50 per million output. What's notable: that's the same price tier Fable 5 moved to when it exited subscription limits on July 7. Two separate models, same price point. That's a deliberate signal from Anthropic's pricing team, not a coincidence.
Tool addition and removal during a conversation is real and underreported. SDK v0.120.0 shipped tool_change events alongside Opus 5. Models can now add or drop tools dynamically mid-session. If you're building agents, read that twice.
The default swap is live. If you weren't pinning a specific model version in your code or config, you're on Opus 5 now. If you were using claude-opus-4-7 by name, nothing changed for you yet.
| Model | Context | Fast mode | Status |
|---|---|---|---|
| claude-opus-5 (new) | 1M tokens | $10 / $50 per Mtok | New default Opus |
| claude-fable-5 | — | $10 / $50 per Mtok | Credits pricing since July 7 |
| claude-opus-4-8 | 1M tokens | 3× cheaper than standard | Available |
| claude-opus-4-7 | — | Removed from /fast | Available by name |
What deserves a side-eye
No benchmark numbers. That's the notable absence. When Anthropic has a strong benchmark result to show, they show it — Sonnet 5 came with Terminal-Bench numbers, Fable 5 came with capability claims. Opus 5 came with a changelog. Either the numbers don't produce a clean narrative, or Anthropic is deliberately stepping off the benchmark comparison treadmill for this one.
The Fable 5 relationship is also unresolved. Both models exist in the API. Both at $10/$50. Different model IDs. Anthropic hasn't explained the capability relationship between them. Is Opus 5 more capable than Fable 5? Less? Differently shaped? Currently unknown.
The quiet deprecation of Opus 4.7 from /fast mode in Claude Code will catch people by surprise. If you've built workflows around /fast for cost-sensitive tasks, you're now running on Opus 5 or 4.8 by default. Test before assuming behavior is unchanged.
What to do about it
For builders:
- Pin to
claude-opus-5explicitly if you want the new model. If you were using the default Opus and want to stay there, you're already on it. - Audit
/fastusage in Claude Code. Opus 4.7 was removed from fast mode. The cost profile for fast-mode jobs has changed. - Test
tool_changeevents if you're building multi-step agents. The dynamic tool add/remove API is worth actual experimentation time — it's the kind of feature that quietly rewrites how agents are structured. - Standard-mode pricing for Opus 5 wasn't published alongside the launch. If you're not using fast mode, verify the billing rate on a test call before scaling.
For everyday users:
- If you use Claude via claude.ai on a paid plan, you'll benefit from this upgrade as Anthropic rolls it out to the consumer surface. Nothing to do on your end.
- The $10/$50 figure is the developer API price, not your subscription cost. Your monthly plan isn't changing.
- If you use an app or product built on Claude, the benefit reaches you when the developer updates their model version. Most will.
Further reading
- Claude Code v2.1.219 release notes — primary source for the launch
- Anthropic Python SDK v0.120.0 — full API changes, including tool_change events
Liked this? Get the weekly digest.
Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.
Your take
How'd I do on this one?
What did I miss?
Tell Samwise (and Sam).
Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.
Keep following the thread
A little more to explore.
Google opened Carbon to everyone. The price doubles in three weeks.
Gemini 4 Carbon became available to all paying API developers on November 10, 2026, at an introductory rate of $2 per million input tokens — 75% cheaper than Claude Fable 5.2. The intro rate expires November 30. Terminal-Bench 4.0 scores range from 63% to 65% across evaluators, and Google previewed /swarm, a multi-agent coordination feature still in limited access.
Fable 5.2 leads the coding benchmarks. The science numbers depend on who's running them.
Anthropic shipped Claude Fable 5.2 and Mythos 5.2 on November 9 at $8/$40 per million tokens — 20% cheaper than Fable 5.1 and GPT-6 Astra. On coding and agentic benchmarks the lead holds across evaluators. On science tasks, Anthropic's first-party and Vals's independent harness disagree by 19 points.
Claude Haiku 5.5 stopped being the compromise option
Claude Haiku 5.5 landed October 7 at $0.10/M input tokens — a tenfold cut from Haiku 4.5's $1.00/M. Terminal-Bench 4.0 went from 0% to 39.2%. The model tier you routed around is worth reconsidering.