Claude Opus 5 vs Fable 5: Which to Use (2026)

TL;DR

• The headline: Claude Opus 5 delivers Fable 5-tier intelligence at half the price, $5 / $25 per million tokens versus Fable's $10 / $50.
• Benchmarks: effectively tied on overall intelligence (AA Index 61 vs 60), and Opus 5 actually beats Fable 5 on 5 of 9 directly comparable tests.
• Where Fable still wins: a narrow set. DeepSWE, the held-out legal benchmark, and the very hardest long-horizon runs where the last percent matters.
• Same specs: both are 1M context, 128K max output. Opus 5's knowledge cutoff (May 2026) is the newer one.
• The call: make Opus 5 your default. Keep Fable 5 for the specific frontier tasks that justify paying 2x.

Anthropic shipped Claude Opus 5 on July 24, 2026, and the pitch is blunt: Fable 5's intelligence for half the money. That's a big claim, because Fable 5 has been the top tier. Here's whether it holds up, benchmark by benchmark, and when you'd still reach for Fable.

What is Claude Opus 5?

Opus 5 is Anthropic's new flagship-value model, built to land Fable 5-class capability at the old Opus price. It keeps the $5 input / $25 output per million tokens that Opus 4.8 charged, so there's no price bump for a real jump in ability. The model id is claude-opus-5. Context is 1M tokens with 128K max output, and the knowledge cutoff is May 2026, the most recent of any Claude model.

The framing Anthropic leans on is "frontier-class agentic coding and computer use at unchanged Opus pricing." In plain terms: the smart tier got cheaper, not the cheap tier got smarter. So what does that buy you against Fable 5? Mostly the same ceiling, minus the premium, which is why this launch matters more than a usual point release.

Opus 5 vs Fable 5: the numbers

Here's the head-to-head on the specs and prices that decide which one you run. Prices are per million tokens.

The story is right there in the first two rows. Opus 5 is exactly half of Fable 5 on both input and output, and the intelligence scores are a dead heat. Almost everything else in the table is identical.

Is it really Fable 5 intelligence for half the price?

Close enough that "half price, same tier" is fair. On the Artificial Analysis Intelligence Index, Opus 5 scores 61 to Fable 5's 60, so they're effectively tied at the top. Independent analysis puts Opus 5's cost-per-task around 26% lower than Fable's while landing the same class of result.

But "tied on average" hides the interesting part. The two trade wins by task. Opus 5 doesn't just match Fable 5. On several benchmarks it beats the more expensive model outright, which flips the usual "you get what you pay for" logic on its head.

One honest caveat before the scores below. Most of these numbers are Anthropic's own announced benchmarks, and independent reproduction is still catching up in the days after launch. Treat them as "the vendor reports," not gospel, and expect third-party testing to sand off some edges. The price gap, though, is a fact, not a claim.

What does the price gap cost at scale?

Half price stops being abstract once you run the volume. Take a steady agentic workload of 1M input and 1M output tokens a day, and the monthly bills split hard:

• Claude Opus 5: about $900 a month
• Claude Fable 5: about $1,800 a month

That's $900 a month, or over $10,000 a year, for the same token volume, before you touch any discounts. Scale it to a busy agent fleet doing ten times that, and the gap turns into six figures a year, which is when finance starts asking why you're on the top tier at all. Opus 5's batch tier drops the bill to roughly $450 a month for non-urgent jobs, and cached reads at $0.50 shave more off repeat-context work like RAG. So the real question was never "which is smarter." It's "is the frontier task in front of me worth 2x," and for most traffic the answer is no.

Where does Opus 5 beat Fable 5?

It leads on most of the agentic and knowledge-work benchmarks, which is exactly where production teams spend money. Opus 5 comes out ahead on Frontier-Bench, GDPval-AA, BrowseComp, OSWorld 2.0, and AutomationBench. Across the nine directly comparable percentage benchmarks, Opus 5 wins five.

Translate that to real work. Opus 5 is stronger on agentic terminal coding, web search and browsing, computer use, and business-workflow automation. If your spend is on coding agents, research agents, or anything driving a browser or terminal, the cheaper model is also the better one here. That's the rare launch where you don't have to choose between price and quality.

Where does Fable 5 still win?

On a narrow, specialist set, plus the hardest long-horizon runs. Fable 5 edges Opus 5 on DeepSWE v1.1 (69.7% to 68.8%) and on the held-out Legal Agent Benchmark (13.3% to 11.7%). Its SWE-Bench Pro sits at 80.3%, and it still ranks near the top of the public leaderboard.

Fable 5's real home is the extreme end: large codebase migrations, long-horizon autonomous coding that runs for hours, senior-grade legal and financial reasoning, and heavy vision work like rebuilding an app from a screenshot. There's also a capability Anthropic deliberately held back on Opus 5. It isn't state-of-the-art for risky dual-use areas like offensive cybersecurity, where Fable's tuning goes further. For most teams that's a non-issue, but it's a real line between the two.

One more difference worth knowing: transparency. Fable 5 ships with safety classifiers that quietly redirect high-risk queries to Opus 4.8, and its raw reasoning is never returned. Opus 5 is the more predictable model to build on when you want consistent behavior without a hidden re-route.

How does Opus 5 compare to Opus 4.8?

This is the easy upgrade, because the price didn't move. Opus 5 charges the exact $5 / $25 that Opus 4.8 did, but it more than doubles 4.8's score on Frontier-Bench and posts clear gains on agentic coding, browsing, and computer use. Same 1M context, same 128K output, and a knowledge cutoff that jumps forward to May 2026.

Is there any reason to stay on 4.8? Rarely. The model id is a drop-in swap from claude-opus-4-8 to claude-opus-5, the price is identical, and the capability is a real step up rather than a rounding error. Unless you've pinned a version for reproducibility, moving up is close to free. Opus 4.8 does keep one quiet role, though: it's the model Fable 5 silently falls back to for high-risk queries, so you may already be using it without asking.

Where does Opus 5 sit in Claude's lineup?

It's the new default for serious work, one rung below the specialist top. Picture four tiers. Fable 5 is the frontier specialist you escalate to. Opus 5 is the high-intelligence default that now covers most of what you'd have paid Fable for. Sonnet 5 sits below for cheaper mid-tier work, and Haiku 4.5 handles high-volume, latency-sensitive jobs. What changed with this launch is that the gap between "default" and "frontier" shrank to almost nothing on both price and capability, so fewer workloads need the top tier at all. If a cheaper model already dominates your bill, none of this moves your setup. For anyone running Opus or Fable, though, Opus 5 resets where the sensible default lives.

So is Fable 5 still worth 2x?

For most workloads, no, and that's the honest answer. If Opus 5 ties or beats Fable on the tasks you actually run, paying double is just paying double. The cases where Fable earns its price are specific. You're pushing the absolute frontier of a long agentic run, you need its edge on a niche benchmark like DeepSWE, or you're in a domain where its extra tuning matters.

The smarter setup for a lot of teams is both, routed by task. Send the bulk of traffic to Opus 5 and escalate only the hardest jobs to Fable 5. You cap cost on the 90% of work that doesn't need the top tier, and still reach for it when a task truly does.

Which should you use?

So which do you actually reach for day to day? For nearly all of it, the cheaper one. Default to Opus 5, escalate to Fable 5 by exception:

• Use Claude Opus 5 for almost everything: coding agents, research and browsing agents, computer use, business automation, and general high-intelligence work. Same tier, half the price, newer knowledge.
• Use Claude Fable 5 for the hardest long-horizon runs, large migrations, niche strengths like DeepSWE or specialized legal reasoning, and cases that need its extra frontier tuning.
• Use both when your workload has a clear split. Route the volume to Opus 5 and escalate only the top few percent to Fable.

Running both means two model ids and, if you're multi-provider, more plumbing. A single OpenAI-compatible endpoint like Velokey lets you switch between Opus 5, Fable 5, and non-Claude models on one key, so task-based routing is a parameter change instead of a new integration. For the numbers behind this, our Claude API pricing guide tracks every tier, the Claude Fable 5 API walkthrough covers the top tier in depth, Claude Sonnet 5 is the step down for cheaper work, and our Gemini 3.6 Flash vs GPT-5.6 vs Claude comparison maps the cheap-and-fast tier if that's more your speed.

Frequently Asked Questions

How much does Claude Opus 5 cost?

Claude Opus 5 is $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8 and half of Fable 5's $10 / $50. Batch processing halves that to $2.50 / $12.50, cached input reads drop to $0.50, and a 2.5x fast mode runs $10 / $50. At a steady 1M-in / 1M-out daily load, that base pricing lands roughly $900 a month under Fable 5.

Is Claude Opus 5 better than Fable 5?

On most benchmarks, yes. Opus 5 ties Fable 5 on overall intelligence and beats it on five of nine directly comparable tests, including agentic coding and computer use. Fable 5 keeps a narrow edge on a few specialist benchmarks and the very hardest long-horizon tasks.

What's the context window for Claude Opus 5?

Claude Opus 5 has a 1M-token context window and a 128K maximum output, matching Fable 5. Its knowledge cutoff is May 2026, the most recent of any Claude model, so it carries fresher training data than Fable 5 does.

Should I switch from Opus 4.8 to Opus 5?

Almost certainly. Opus 5 keeps Opus 4.8's exact pricing but more than doubles its score on Frontier-Bench and posts big gains on agentic and computer-use tasks. Same cost, meaningfully more capable, so there's little reason to stay on 4.8.

When is Fable 5 still worth the higher price?

When you're at the frontier: hours-long autonomous agent runs, large codebase migrations, or niche strengths like DeepSWE and specialized legal reasoning where Fable still leads. For everyday high-intelligence work, Opus 5 delivers the same tier for half the cost.

What is the Claude Opus 5 model id?

The model id is claude-opus-5 on the Claude API and anthropic.claude-opus-5 on Amazon Bedrock. It's a drop-in swap for claude-opus-4-8 at the same price, so most existing Opus integrations only need the id changed.

Is Claude Opus 5 available on the API now?

Yes. Opus 5 is live on the Claude API as claude-opus-5 and on Amazon Bedrock as anthropic.claude-opus-5, at the same $5 / $25 pricing as Opus 4.8. Batch processing, prompt caching, and a 2.5x fast mode are all available, so you can tune cost and speed per workload.