Claude Fable 5.1 vs Opus 5: When Is Twice the Price Worth It?

Claude Fable 5.1 vs. Claude Opus 5: Which Model Should You Use?

Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens. Claude Opus 5 costs $5 and $25. Anthropic recommends starting with Opus 5 and switching to Fable 5.1 only “when your evals on Claude Opus 5 at higher effort still fall short.” This comparison applies that rule to benchmarks, pricing, caching, and real-world integration constraints.

Try Apidog today

The sources are Anthropic’s launch post, the Fable 5.1 model page, and the pricing page. For each model’s overview, see what Claude Fable 5.1 is and what Claude Opus 5 is.

Side by side

	Claude Fable 5.1	Claude Opus 5
Model ID	claude-fable-5-1	claude-opus-5
Released	September 1, 2026	July 24, 2026
Positioning	Demanding reasoning and long-horizon agentic work	Complex agentic coding and enterprise work; the recommended default
Input / output	$10 / $50	$5 / $25
Cache read	$0.25	$0.50
Cache write (5m / 1h)	$12.50 / $20	$6.25 / $10
Batch	$5 / $25	$2.50 / $12.50
Fast mode	No	Yes ($10 / $50)
Priority Tier	No	No
Context / max output	1M / 128K	1M / 128K (300K on Batch with beta)
Knowledge cutoff	June 2026	May 2026
Comparative latency	Slower	Moderate
Thinking	Always on; disabled returns 400	On by default; disabled allowed at high or below
Forced tool_choice	400	Accepted
Safety classifiers	cyber, bio, frontier_llm, reasoning_extraction, general_harms	cyber only
Zero data retention	Not available unless authorized	Available
Per-message effort (beta)	Yes	Yes
History-editing check	Yes	No

What the benchmarks show

All results below come from Anthropic, were run by Anthropic, and have not yet been independently reproduced.

Benchmark	Fable 5.1	Opus 5	Gap
Terminal-Bench-Science 0.1	52.6%	29.0%	+23.6
AutomationBench	31.4%	26.9%	+4.5
Humanity’s Last Exam (no tools)	60.9%	56.6%	+4.3
Terminal-Bench 4.0	55.8%	52.3%	+3.5
CursorBench 3.2.0	73.4%	70.0%	+3.4
OSWorld 2.0 (partial)	77.9%	75.4%	+2.5
OSWorld 2.0 (strict)	41.7%	39.6%	+2.1
Humanity’s Last Exam (with tools)	65.0%	63.6%	+1.4
GDPval-AA v2	1853	1824	+29 Elo

The results tell two different stories:

• On Terminal-Bench-Science, Fable 5.1 nearly doubles Opus 5. This is a long-horizon, tool-heavy research benchmark where models run terminal experiments over extended periods.
• On the other benchmarks, the advantage ranges from one to five points. A five-point lead on AutomationBench is meaningful, while a 1.4-point lead on Humanity’s Last Exam with tools—or 29 Elo on GDPval—may be smaller than the effect of prompt design and effort settings.

Opus 5 launched in July at 52.3% on Terminal-Bench 4.0, ahead of Fable 5’s 42.0%. Fable 5.1 retook the lead at 55.8%, but the current difference is 3.5 points rather than 13.8. The Fable tier’s advantage in agentic coding is real, but narrower than it was in June. The Opus 5 vs. Fable 5 comparison covers the earlier results.

The pricing math is not simply 2×

Without caching, Fable 5.1 costs exactly twice as much as Opus 5 on every pricing line. Cache reads reverse that relationship: Fable 5.1 costs $0.25, compared with $0.50 for Opus 5.

The effective multiplier therefore depends on how much of each request is a cached prefix.

Consider an agentic turn with 150,000 cached tokens, 1,000 uncached input tokens, and 800 output tokens:

• Opus 5: 5(0.001) + 0.5(0.15) + 25(0.0008) = $0.005 + $0.075 + $0.02 = $0.10
• Fable 5.1: 10(0.001) + 0.25(0.15) + 50(0.0008) = $0.01 + $0.0375 + $0.04 = $0.0875

On this turn, Fable 5.1 is cheaper.

Now consider a code review with 30,000 uncached input tokens and 4,000 output tokens:

• Opus 5: $0.15 + $0.10 = $0.25
• Fable 5.1: $0.30 + $0.20 = $0.50

Here, Fable 5.1 costs exactly twice as much.

Fable 5.1’s per-turn cost falls below Opus 5’s when cached input tokens exceed roughly 20 times the sum of uncached input tokens and five times output tokens. This is common in prefix-heavy agent loops but impossible for output-heavy workloads. Cache writes also cost twice as much on Fable 5.1, so frequently resetting a cache can eliminate the advantage.

See the Fable 5.1 pricing breakdown and the Opus 5 pricing breakdown for the complete tables.

Where Opus 5 wins

Workloads that can run with thinking disabled. Opus 5 accepts thinking: {"type": "disabled"} at high effort or below. Fable 5.1 does not. For classification, extraction, and short-answer routes, Opus 5 with thinking disabled—or at low effort—is cheaper and faster.

Fast mode. Opus 5 offers a research-preview fast mode with up to 2.5× the output speed for $10 per million input tokens and $50 per million output tokens. Fable 5.1 has no equivalent.

Latency. Anthropic lists Fable 5.1 as “slower” and Opus 5 as “moderate.” A difficult Fable 5.1 request at high effort can run for several minutes.

Zero data retention. Opus 5 is available under ZDR. Fable 5.1 is a Covered Model and returns a 400 unless Anthropic expressly authorizes ZDR access.

Fewer refusals. Opus 5 uses cybersecurity-only classifiers. Fable 5.1 also applies bio, frontier LLM development, reasoning extraction, and general-harms classifiers. A benign life-sciences or machine-learning workload may pass on Opus 5 but trigger Fable 5.1. When that happens, Opus 5 is the fallback target.

Forced tool use. Opus 5 accepts tool_choice values any and tool; Fable 5.1 returns a 400. Integrations that depend on forced calls may need changes before switching.

History editing. Opus 5 permits edits to earlier turns. Fable 5.1 invalidates later thinking blocks when earlier history changes. Client-side compaction and per-turn reminder injection therefore require an audit before migration.

Where Fable 5.1 wins

Long-horizon research and terminal agents. The 52.6% versus 29.0% result on Terminal-Bench-Science is the largest gap in Anthropic’s table. This is the workload Fable 5.1 was designed for: running experiments, conducting multistep web research, and operating for hours.

Business workflow automation. Fable 5.1 scores 31.4% versus Opus 5’s 26.9% on AutomationBench, which models end-to-end tasks across applications.

Cached-prefix-heavy agents. The $0.25 cache-read price can make Fable 5.1 cheaper per turn when most input tokens come from a reusable prefix.

Knowledge cutoff. Fable 5.1’s June 2026 cutoff is one month newer than Opus 5’s May 2026 cutoff.

Dense-document vision. Anthropic identifies improved reading of charts, filings, and tables embedded in PDFs—especially with crop-and-zoom tools—as a Fable 5.1 strength. Opus 5 has not been compared on a published vision benchmark, so test this claim on your own documents.

Tasks Opus 5 fails at xhigh. This is Anthropic’s recommendation: if a task fails your Opus 5 eval at xhigh, test Fable 5.1 at high before changing anything else.

Effort changes the comparison

Both models provide five effort levels. Anthropic specifically claims that Fable 5.1 at low effort is often competitive with Claude Opus and Claude Sonnet models on cost per task while scoring higher.

That makes the practical comparison more nuanced than Opus 5 at high versus Fable 5.1 at high. For routine work, compare Opus 5 at high with Fable 5.1 at low or medium. Lower effort usually produces fewer output tokens, offsetting Fable 5.1’s higher per-token price.

The reverse is also true: Fable 5.1’s gains over Fable 5 are largest at xhigh and max, where latency and cost are highest. Sweep both models across all relevant effort levels on your own evals before relying on the launch tables. The Opus 5 effort guide covers all five levels; the semantics are the same for Fable 5.1.

A practical decision framework

1. Start with Opus 5 at high. It is Anthropic’s default recommendation, costs half as much, uses fewer classifiers, and supports ZDR.
2. Raise failing tasks to xhigh. This is cheaper than switching models for every request.
3. Move only the remaining failures to Fable 5.1 at high. Use a per-route policy rather than switching globally.
4. Check the cache profile. If the migrated route uses a large reusable prefix, Fable 5.1 may cost less per turn.
5. Audit your harness before migrating. Forced tool_choice and history edits both break on Fable 5.1. Use the migration guide as a checklist.

Run the comparison in Apidog

Create one collection containing your production prompts, with model configured as an environment variable. Run the collection with:

• claude-opus-5 at high
• claude-opus-5 at xhigh
• claude-fable-5-1 at high
• claude-fable-5-1 at low

For every run, record the response, usage.output_tokens, and usage.cache_read_input_tokens on a repeat request. Apply each model’s rates in a post-response script to calculate cost per task, model, and effort level in one table.

Apidog keeps every run in history, so you can repeat the comparison when the next model launches. Download Apidog to get started.

FAQ

Is Claude Fable 5.1 better than Opus 5?

On every benchmark Anthropic published, yes, with margins ranging from 1.4 points to 23.6 points. The largest gap is on Terminal-Bench-Science; most other results are under five points. These are vendor-run results.

Is Fable 5.1 worth twice the price of Opus 5?

Often, for long-horizon research, terminal agents, business automation, and prefix-heavy loops where cache reads can reduce the per-turn cost. Usually not for output-heavy, latency-sensitive, or classification workloads. Anthropic’s rule is to switch only after Opus 5 at higher effort fails your evals.

Which is faster, Fable 5.1 or Opus 5?

Opus 5. Anthropic lists its latency as “moderate” versus “slower” for Fable 5.1. Opus 5 also offers a fast mode with up to 2.5× the output speed; Fable 5.1 does not.

Does Opus 5 have the same safety classifiers as Fable 5.1?

No. Opus 5 uses cyber-only classifiers. Fable 5.1 also applies bio, frontier LLM, reasoning extraction, and general-harms classifiers. When Fable 5.1 refuses a request, Opus 5 is one of the two permitted fallback targets.

Can I use Fable 5.1 under zero data retention like Opus 5?

No. Fable 5.1 requires 30-day retention and is a Covered Model. Opus 5 is available under ZDR.

Can a conversation move between Opus 5 and Fable 5.1?

Upward, yes: Fable 5.1 reads Opus 5’s thinking blocks. Downward, the API removes Fable 5.1’s thinking blocks before Opus 5 receives them. Those blocks are unbilled, and Opus 5 replans the turn.