Opus 5 vs Fable 5 is a choice between two tiers of Anthropic’s own lineup: the new near-flagship workhorse and the top-tier flagship above it. Opus 5, released July 24, 2026, is the everyday model, priced at half of Fable 5 and, on Anthropic’s own evaluations, matching or beating Fable 5 on a lot of coding and knowledge work. Fable 5 is still the top of the range, the model built for the genuinely hardest frontier tasks. Because both come from the same vendor and are current, you can compare them head to head today, and the useful question is not which is better overall but which fits the task in front of you.
The short version: route most work to Opus 5. It costs $5 and $25 per million input and output tokens against Fable 5’s $10 and $50, and Anthropic says it delivers near-flagship performance, even leading on some coding and knowledge-work evals, at roughly half the price. Fable 5 stays worthwhile for the hardest, highest-stakes problems where the top tier still earns its premium. This guide places both models, shows the price and benchmark gaps, and gives a plain rule for choosing, plus how to use Opus 5’s effort setting to keep spend down. Treat the head-to-head benchmark claims as Anthropic’s framing until independent tests land.
Opus 5 vs Fable 5 at a glance
Opus 5 and Fable 5 sit one rung apart. Opus 5 is the high-capability workhorse tier, the model Anthropic wants most enterprises, developers, and knowledge workers to reach for by default. Fable 5 is the public, safeguarded configuration of the top Mythos-class tier, the flagship reserved for frontier-hard work; Mythos 5 is its trusted-access sibling. For more on where Opus 5 landed, see our explainer on Claude Opus 5, and for the other end of the family, our Sonnet 5 vs Fable 5 guide.
Availability is part of the picture too. Anthropic has made Opus 5 the new default model on Claude Max and the strongest model available on Claude Pro, which means many subscribers now meet Opus 5 without choosing it. Fable 5, by contrast, settled into a permanent slot on Max and Team Premium after a turbulent few months.
One thing that does not separate them is context. Both offer a 1M-token context window, and Opus 5 can produce up to 128K output tokens in a single pass, so how much you can feed the model or get back is not the deciding factor. What drives the decision is price, capability on the hardest tasks, and how much you trust a vendor’s own benchmarks.
What each model costs
Opus 5 costs $5 per million input tokens and $25 per million output tokens, the same rates as the Opus 4.8 it replaces. Fable 5 sits at $10 and $50, exactly double on both sides. Because output tokens dominate most real bills, the output column is the one to watch: $25 against $50 per million is a straight halving. At any real volume, that gap shapes architecture, not just budget. If the cost math is new to you, our explainer on input versus output tokens walks through why output usually drives the invoice.
What Anthropic claims about performance
Price only matters next to capability, and this is where Opus 5 gets interesting. Anthropic positions Opus 5 as its best-performing everyday model and says it outperforms Fable 5 on coding and knowledge-work evaluations while costing half as much.
On Frontier-Bench v0.1, an agentic terminal coding benchmark run independently by Harbor and the Laude Institute, Anthropic reports Opus 5 at 43.3%, ahead of Fable 5’s 33.7% and more than double Opus 4.8’s 18.7%. On CursorBench 3.2, Anthropic says Opus 5 at maximum effort lands within half a percent of Fable 5’s peak while costing roughly half as much per task. An important distinction: the benchmarks themselves are third-party, but these particular figures come from Anthropic’s own internal runs, which its footnotes describe as the mini-SWE-agent harness on a GKE backend averaged over five attempts per task. That is more credible than a purely in-house eval and still short of independent verification, so read them as framing rather than a verdict. The direction is clear enough, though: on everyday coding and knowledge work, Opus 5 is at least competitive with the flagship for far less money.
Where Fable 5 still earns its premium
The head-to-head evals do not make Fable 5 pointless, but it is worth being precise about where the remaining gap actually is, because Anthropic’s own framing is narrower than the marketing shorthand suggests. Anthropic describes Opus 5 as coming close to Fable 5’s frontier intelligence at half the price and calls it the new state of the art on coding and knowledge-work evaluations. It does not publish a claim that Fable 5 beats Opus 5 on general hard tasks. The clearly documented limitation is narrower and specific: Anthropic states that Opus 5 remains behind Mythos 5, the trusted-access sibling at the top of the range, on cybersecurity work, and that while Opus 5 comes close at finding vulnerabilities it is substantially behind at developing exploits. There is also a practical access difference. Opus 5 ships with cyber classifiers that Anthropic says are far less restrictive than Fable 5’s, and flagged requests fall back to Opus 4.8 by default in Claude.ai, Claude Code, and Claude Cowork. So the honest read is this. Reaching for Fable 5 still makes sense on frontier, research-style reasoning, the most demanding long-horizon agentic chains, and high-stakes output where a marginally better answer justifies paying twice as much, and that judgment rests partly on inference from where the top tier sits rather than on a published head-to-head loss for Opus 5. Verify it on your own hardest tasks rather than assuming it. Fable 5 also had a bumpy road to get here, going offline in June before returning and settling into a permanent slot in Max and Team Premium; our piece on Fable 5’s usage-limits saga covers how access finally stabilized. The point is not that Opus 5 wins everywhere. It is that the cases where Fable 5 clearly wins are narrower than they used to be.
When to use Opus 5
Default to Opus 5 and escalate only when you have a concrete reason. It is the right choice for:
- Everyday coding. Feature work, refactors, tests, and reviews, including much of what used to feel like flagship territory.
- Knowledge work and analysis. Research, writing, summarizing, and structured reasoning, where Anthropic’s evals put it at or above Fable 5.
- High-volume or user-facing features. Anything where per-call cost compounds, since Opus 5 is half the price of the flagship.
- Agent loops with many calls. When a task fans out into dozens or hundreds of calls, the cost gap multiplies in your favor.
- Long-context jobs. Holding a large codebase or a stack of documents in one 1M-token session and producing long output in a single pass.
When to use Fable 5
Reach for Fable 5 when the task is genuinely at the frontier and the stakes justify the premium:
- The hardest agentic work. Long-horizon, multi-file, many-step tasks where you can see Opus 5 stall rather than merely produce a rougher answer.
- Frontier reasoning. Novel, research-style problems that sit at the edge of what current models can do.
- High-stakes output. Cases where a marginally better result clearly justifies paying twice as much per call.
- Verification against the top tier. When you want the flagship’s answer as a check on a critical Opus 5 result.
If you are weighing the whole Claude range rather than just these two, our guide to picking between Opus, Sonnet, and Haiku covers the tiers below the flagship.
Using the effort setting to control cost
Opus 5’s most practical lever is its five-level effort setting. Instead of a fixed reasoning budget, you choose how hard the model thinks, from a quick low-effort pass to a maximum reasoning mode, which directly controls both latency and spend. For routine work you leave effort low and pay for a fast, cheap response; for a thorny bug or a dense analysis you turn it up and get more deliberation on the same $5 and $25 pricing. That dial changes the comparison with Fable 5, because a lot of the tasks that once justified reaching for the flagship can now be handled by turning Opus 5’s effort up rather than paying double. Use the setting deliberately: default low, escalate per task, and reserve Fable 5 for the cases where even high-effort Opus 5 is not enough.
How to decide
Start every task on Opus 5. If it falls short, first try raising the effort setting before you reach for a more expensive model, and escalate to Fable 5 only when the work is clearly frontier-level and high-stakes. Let the workload pick the model rather than routing everything through the top tier by reflex, because the price gap is large and, on Anthropic’s own evals, the everyday quality gap is small or reversed. The competitive backdrop reinforces the same habit: cheaper near-frontier models are arriving from every direction, including OpenAI’s GPT-5.6, and staying model-agnostic keeps you free to route each task to whatever gives the best result per dollar.
The bottom line on Opus 5 vs Fable 5: it is rarely an all-or-nothing choice for a whole system. Opus 5 is the right default for the large majority of coding and knowledge work at half the price, and Fable 5 is the specialist you route the hardest, highest-stakes tasks to. Match the model to the task, lean on the effort setting, and you capture most of the flagship’s value for a fraction of the cost.
Frequently Asked Questions
What is the main difference between Opus 5 and Fable 5?
Opus 5 is Anthropic’s near-flagship workhorse, released July 24, 2026, priced at half of Fable 5 and meant for most everyday coding and knowledge work. Fable 5 is the top-tier flagship above it, built for the hardest frontier tasks. In short, Opus 5 is the sensible default and Fable 5 is the specialist you reserve for the toughest, highest-stakes work.
Is Opus 5 really better than Fable 5?
On Anthropic’s own coding and knowledge-work evaluations, Opus 5 matches or beats Fable 5 at roughly half the cost. Those are the vendor’s numbers on the vendor’s harness, so treat them as framing until independent benchmarks confirm them. Fable 5 still leads on the hardest, longest-horizon problems, so which is better depends on the task.
How much cheaper is Opus 5 than Fable 5?
Exactly half on both sides. Opus 5 is $5 per million input tokens and $25 per million output, while Fable 5 is $10 and $50. Since output usually dominates a bill, the practical saving at real volume is close to that full halving.
Do Opus 5 and Fable 5 have the same context window?
Yes. Both offer a 1M-token context window, and Opus 5 can return up to 128K output tokens in a single response. How much you can feed the model or get back is not what separates them; price and capability on the hardest tasks are.
When is Fable 5 still worth the premium?
When the task is genuinely at the frontier: long-horizon multi-file agentic work where Opus 5 stalls, novel research-style reasoning, or high-stakes output where a marginally better answer justifies paying twice as much. For routine coding, analysis, and writing, Opus 5 is the more economical choice.
What does Opus 5’s effort setting do?
It lets you choose how hard the model thinks across five levels, from a quick low-effort pass to a maximum reasoning mode. Effort is a direct lever on latency and cost, so you keep it low for routine work and turn it up for hard problems, often closing the gap to Fable 5 without paying the flagship’s price.
Should I switch from Fable 5 to Opus 5?
For most workloads, it is worth testing. Route your everyday coding and knowledge work to Opus 5, keep Fable 5 for the genuinely hardest cases, and re-run your own evaluations before assuming Anthropic’s benchmark gains show up in your specific tasks. At half the price, the downside of trying Opus 5 is small.
