Anthropic spent two months telling the world their most powerful AI model was too dangerous to release. Yesterday they released it anyway. I have been writing about this model since April, and I finally got my hands on it. Here is what actually happened.
A Model Two Months in the Making
If you have been following along here, none of this came out of nowhere. Back in April I wrote about Claude Mythos, the model Anthropic had quietly built and then quietly panicked about. First there was the access question, whether a model this expensive would ever reach everyday people. Then came Project Glasswing, the initiative where Anthropic gave Mythos to a handful of infrastructure companies and governments before anyone else, because they genuinely believed releasing it carelessly could make large-scale cyberattacks more likely.
That model is now public. Sort of.
On June 9, 2026, Anthropic launched Claude Fable 5, the first publicly available model in the Mythos class. They also launched Claude Mythos 5 on the same day, but that one remains restricted to a small group of security and infrastructure organisations through Project Glasswing. Fable 5 is the version the rest of us get. Same family, different safeguards. In high-risk areas like cybersecurity, biology, and chemistry, the model routes your query to Claude Opus 4.8 instead of answering directly. Anthropic says this fallback affects fewer than 5% of sessions, so for most use cases you are getting the full thing.
The name is intentional. Fable comes from the Latin fabula, meaning “that which is told,” which is a deliberate nod to Mythos. Anthropic does love a theme.
What Makes It Different
The headline capability is autonomy. Anthropic says Fable 5 can work on long, complex tasks more independently than any previous Claude model. For software engineers specifically, they ran an internal test where a Ruby codebase with 50 million lines of code was migrated in a single day. That same task would have taken an entire team over two months to complete by hand.
It also has a 1 million token context window, 128,000 maximum output tokens, and a knowledge cutoff of January 2026. Benchmark scores put it above Opus 4.8 across coding, vision, knowledge work, and scientific research. The longer and more complex the task, the wider the gap between Fable 5 and everything below it.
I opened Claude desktop on my Mac this morning and the banner was right there. Fable 5, available now. I switched models and got to work.
I Put It on My Game
I have been building a car racing game called CHOP FIRST using vibe coding, mainly through Codex GPT 5.5. If you have never heard of vibe coding, the short version is that you describe what you want to an AI and it writes the code. You test, you break things, you describe the problem, it tries again. Repeat until something works.
CHOP FIRST has been in that loop for a while. Drift physics that felt wrong. The car getting glued to the road rails instead of bouncing off them naturally. Reversing that behaved like the laws of physics had taken the day off. Codex GPT 5.5 was making progress but it was slow, iterative, and each fix had a habit of breaking something else. The kind of grind that makes you question whether the game will ever feel right.
Before switching to Fable 5, I had Codex draft me a comprehensive prompt summarising everything it had built, every issue still persisting, the goal of the game, and the phases still left to complete. I fed that entire brief to Fable 5 in a single prompt.
The far camera view was upgraded. The steering felt noticeably tighter and more responsive. The environment and road rendering improved. The physics issues that had been the most stubborn, the gluing, the drift inconsistency, were largely resolved. I would put it at around 80% of the outstanding problems addressed in one shot.
That is not a small thing. That is weeks of back-and-forth compressed into one well-structured prompt.
The Part That Stings
Here is the thing though. Fable 5 is priced at $10 per million input tokens and $50 per million output tokens through the API. That is double what Opus 4.8 costs, and Opus already burns through credits faster than I am comfortable with on a Pro plan. I wrote in April that I was concerned Mythos-class AI would price out everyday builders. Fable 5 is the proof point.
Right now, from June 9 through June 22, Fable 5 is included at no extra cost on Pro, Max, Team, and Enterprise plans. That is the window you are in today if you are reading this fresh. After June 22, it moves behind usage credits. Anthropic says they want to restore it as a standard plan feature as soon as capacity allows, but there is no committed date.
So my honest experience is this. I tested the most impressive AI model I have ever used for a task I actually care about, watched it do in one prompt what months of iteration had not, and then checked the pricing and remembered why my default is still Sonnet 4.6.
That gap, between what frontier AI can demonstrably do and what everyday builders can realistically afford to use on a Tuesday, is the thing worth paying attention to. I said it in April. Fable 5 just made it more visible.
Should You Try It Before June 22?
Yes. If you are on a Pro, Max, or Team plan, you have until June 22 to use Fable 5 at no extra cost. That is 12 days from today. Use it on your hardest problem. Your most stuck project. The thing you have been going back and forth on with a cheaper model for weeks. Give it a well-structured, detailed prompt and see what it does.
Just go in with realistic expectations about what happens on June 23.
Frequently Asked Questions
What is Claude Fable 5 and how is it different from Claude Opus? Claude Fable 5 is Anthropic’s first publicly available Mythos-class model, sitting above the Opus tier in capability. It outperforms Opus 4.8 across coding, vision, knowledge work, and scientific research, particularly on long and complex tasks. The main practical differences are higher output quality on difficult problems, greater autonomy on multi-step tasks, and a significantly higher price point.
Can I use Claude Fable 5 for free on Claude Pro? Yes, but only until June 22, 2026. From launch through June 22, Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. From June 23, using it requires usage credits billed at API rates. Anthropic has said they aim to restore it as a standard plan feature once capacity allows, but has not given a date.
Is Claude Fable 5 good for coding and building projects? Based on my own test, it is noticeably better than previous models on complex, long-horizon coding tasks. It handled physics and environment issues in a game project in a single prompt that had been grinding for weeks on another model. For developers and vibe coders with clearly structured prompts, the results are genuinely impressive. Whether the quality jump justifies the cost after June 22 depends on how often you hit problems that actually need that level of capability.
The future is already here. It is just not evenly distributed. And right now, the distributed part costs $50 per million output tokens.
Leave a Reply