Anthropic dropped its biggest release of the year on September 1, and the naming scheme seems designed to confuse us. Claude 5.1 ships as two models, Fable and Mythos, that share the exact same brain, and the price cut everyone’s talking about has nothing to do with the sticker price. This one’s worth your attention, and not just because the benchmark numbers look absurd.
Why model launches feel like noise
Every couple of weeks, another “best model ever” shows up. The blurbs blur together. ChatGPT gets a new number, Google renames something, a startup claims a new state of the art, and for most people none of it changes what they do on Monday morning. You’re right to be skeptical.
Still, every so often a release changes the economics of using AI, and that’s when it stops being tech gossip and starts touching your wallet. Claude 5.1 is that kind of release. Two things happened here: the model itself genuinely improved, and the cost of keeping an AI working on long tasks cratered. Both matter, just to different people.
Two names, one brain: Fable and Mythos
Here’s the simplest way to think about this launch: Fable and Mythos are the same model running with different safety settings. Same weights, same intelligence, and honestly it’s easier to think of Claude 5.1 as one brain wearing two different seatbelts.
Fable 5.1 is the standard version. It’s the one you get in the Claude apps (Pro, Max, Team, and Enterprise). When a request crosses into genuinely dangerous territory, like cyber or bio help, it gets routed to an Opus model that acts as a second gate.
Mythos 5.1 is the same model with those rails loosened. It’s not for sale to the general public. Access is restricted to US organizations doing things like cyber defense and life-science research, through a trusted-access track run with the US government.
Why should a beginner care about a model they can’t open? Because it shows where the industry is heading. Safety is becoming product architecture instead of a policy document, and Anthropic is the first vendor treating it that openly. You can read the details on Anthropic’s official Fable 5.1 page.
Where Claude 5.1 actually got better
The benchmark table reads like a sports scoreboard, and the AI news briefing from Latent Space did the digging on the details. On Terminal-Bench 4.0, a coding test suite, Claude 5.1 hit 55.8 percent, while Opus 5 managed 52.3 percent and OpenAI’s GPT-5.6 Sol landed at 37.3 percent. That’s a meaningful gap, even if the coding lead is narrower than the raw numbers suggest.
The jumps are bigger elsewhere. On agentic science tasks, it more than doubled the old score: 52.6 percent versus 24.7 percent for Fable 5. Business workflows nearly doubled too, hitting 31.4 percent on AutomationBench.
Underneath the flashy numbers, the knowledge cutoff moved to June 2026, five months fresher than Fable 5. It knows newer stuff, which matters more than people admit.
The underrated part for normal users is the feel. This model keeps working longer before it needs you to jump back in, it tells you when it’s stuck instead of guessing, and its writing got more natural. If you use Claude for everyday work, that last one is the improvement you’ll notice first.
The price story is the cache
Headline pricing didn’t budge: $10 per million input tokens, $50 per million output tokens. But cache reads crashed from $1.00 to $0.25 per million tokens, a 75 percent cut.
What’s a cache read? When you work on a big project, the model re-reads your entire history on every single request. That repeated context is usually the most expensive part of a long session. Caching charges you a tiny fraction for that reused context, which turns expensive agent loops into something you can actually run all day.
Anthropic estimates typical workloads now run about 25 percent cheaper, and highly agentic workloads up to 45 percent cheaper. Cognition, the company behind Devin, said it’s moving Opus 5 traffic to Fable 5.1 on launch day because the new cache price finally makes a top-tier model economical for the workloads it had kept on Opus, as VentureBeat reported.
| Claude model | Input (per 1M) | Cache read (per 1M) | Output (per 1M) |
|---|---|---|---|
| Fable 5.1 | $10 | $0.25 | $50 |
| Fable 5 | $10 | $1.00 | $50 |
| Opus 5 | $5 | $0.50 | $25 |
| Sonnet 5 | $2 | $0.20 | $10 |
For context, Sonnet 5 is the cheap workhorse and Opus is the premium brain. Fable 5.1 costs more per token than Opus, so it only makes sense when the task is big enough to lean on the cache discount. That’s why Anthropic is pitching it at long-running, ambitious projects rather than quick chat.
What Claude 5.1 means for you
If you use the Claude apps, you’re already getting the smarter model. Longer tasks with fewer interruptions, more natural writing, fresher knowledge. You don’t need to do anything.
If you build with the API, this is your signal to switch cache-heavy workloads to claude-fable-5-1. Agent loops, document pipelines, long code sessions: those are exactly the jobs where the 75 percent cache cut pays off. On top of that, US-only inference is available at 1.1x pricing if your work needs to stay stateside.
If you’re on ChatGPT or Gemini, watch what happens next. When one vendor slashes costs this hard, the others tend to follow, and users win either way. Free and cheaper models keep arriving too, like the Qwen 3.8 review we published, and Gemini 3.7 Flash went free recently. The market is moving in one direction.
One honest hot take: if you’d been avoiding Claude because long sessions felt pricey or rate-limited, the new caching math changes the argument. The expensive part of using AI has been keeping it working, not turning it on. That’s the part that just got cheaper.
What didn’t change
Worth saying what stayed the same, because launch day always brings rumors. Claude Pro, Max, Team, and Enterprise subscription prices didn’t move. Opus and Sonnet are still around and still cheaper per token. Batch pricing is unchanged too.
So the list of real changes is short: a smarter model, a fresher knowledge cutoff, and that 75 percent cache cut. That last one is what shakes the economics of long-running AI work, and it’s the number worth remembering.
The bottom line
Claude 5.1 is a real upgrade, not a marketing refresh. Smarter on the tasks that matter, cheaper to run for long work, and available today in the apps you probably already use.
Try it the way you actually work. Give it one long, messy task, the kind you’d normally split into ten separate chats, and let it run. That’s where this model shows its edge, and with the cache price cut, it’s the first time letting an agent work all afternoon doesn’t feel like a luxury. For more context on how model prices ripple through the industry, check out what OpenAI’s own chip means for future pricing.
