minirouter.sh
Open navigation
Get a key

Anthropic

Claude Fable 5.1

Anthropic's most capable generally available model. Same list price as Fable 5, cache reads at a quarter of the cost, biggest gains on long-running agentic work.

successor to claude-fable-5text + image + pdf inputtext output

Replaces Claude Fable 5: same input and output price, same context window.

Get a MiniRouter keySee the evidence

Run this model

anthropic/claude-fable-5.1

const response = await fetch('https://api.minirouter.sh/v1/chat/completions', {  method: 'POST',  headers: {    Authorization: `Bearer ${process.env.MINIROUTER_KEY}`,    'Content-Type': 'application/json',  },  body: JSON.stringify({    model: 'anthropic/claude-fable-5.1',    messages: [{ role: 'user', content: 'Why is the sky blue?' }],  }),}) const data = await response.json()console.log(data.choices[0].message.content)
Read docs →

Model overview

Input / 1M
$10.5
All-in MiniRouter rate
Output / 1M
$52.5
All-in MiniRouter rate
Context
1M
tokens
Max output
128K
tokens
Effort
low → xhigh
adaptive thinking, always on
Cache read
$0.25 list
was $1.00 on Fable 5
Released
31 Aug 2026
publisher release

Publisher-reported, plotted by MiniRouter

The jump is in the long-running work.

Anthropic's launch table puts Fable 5.1 ahead of Fable 5, Opus 5, and GPT-5.6 Sol on all nine evaluations. The gaps are widest on agentic research and business workflows. MiniRouter plotted the table and did not run the evaluations.

Agentic scientific research

More than double Fable 5 on Terminal-Bench-Science.

Multi-step research tasks in a terminal. Standard error is ±3.5 to 4.5 points, so the lead is well outside noise.

Fable 5.1
52.6%
vs Fable 5
2.1×24.7%
vs Opus 5
1.8×29.0%

Terminal-Bench-Science 0.1

0 to 100%

  • Claude Fable 5.152.6%1
  • Claude Fable 524.7%
  • Claude Opus 529.0%
  • GPT-5.6 Sol22.4%
¹ Anthropic's internal harness; the public leaderboard has Opus 5 at 30.0% and Fable 5 at 21.4%. Source: Anthropic announcement ↗.
View chart data
Terminal-Bench-Science 0.1 scores
ModelScoreRelative to Fable 5.1
Claude Fable 5.152.6%—
Claude Fable 524.7%47%
Claude Opus 529.0%55%
GPT-5.6 Sol22.4%43%

Eight more evaluations

Coding, computer use, knowledge work, reasoning

  • Claude Fable 5.1
  • Comparison models

Agentic coding

Terminal-Bench 4.0

  • Claude Fable 5.155.8%
  • Claude Fable 542.0%
  • Claude Opus 552.3%
  • GPT-5.6 Sol37.3%

0 to 100%

Agentic coding

CursorBench 3.2.0

  • Claude Fable 5.173.4%
  • Claude Fable 570.5%
  • Claude Opus 570.0%
  • GPT-5.6 Sol67.2%

0 to 100%

Business workflows

AutomationBench

  • Claude Fable 5.131.4%
  • Claude Fable 517.1%
  • Claude Opus 526.9%
  • GPT-5.6 Sol19.6%

0 to 100%

Knowledge work

GDPval-AA v2 (Elo)

  • Claude Fable 5.11,853
  • Claude Fable 51,723
  • Claude Opus 51,824
  • GPT-5.6 Sol1,711

1,600 to 1,900 Elo

Computer use

OSWorld 2.0 · partial

  • Claude Fable 5.177.9%2
  • Claude Fable 572.9%
  • Claude Opus 575.4%
  • GPT-5.6 Sol—Not reported

0 to 100%

Computer use

OSWorld 2.0 · strict

  • Claude Fable 5.141.7%2
  • Claude Fable 536.1%
  • Claude Opus 539.6%
  • GPT-5.6 Sol—Not reported

0 to 100%

Multidisciplinary reasoning

Humanity's Last Exam · no tools

  • Claude Fable 5.160.9%
  • Claude Fable 557.8%
  • Claude Opus 556.6%
  • GPT-5.6 Sol—Not reported

0 to 100%

Multidisciplinary reasoning

Humanity's Last Exam · with tools

  • Claude Fable 5.165.0%
  • Claude Fable 563.8%
  • Claude Opus 563.6%
  • GPT-5.6 Sol—Not reported

0 to 100%

Run with production safeguards on; tasks they intervened in scored zero. ² OSWorld 2.0 uses the August 2026 task set, so no competitor score is comparable. Source: Anthropic announcement ↗.
View chart data
Complete values from Anthropic's Claude Fable 5.1 launch table
BenchmarkClaude Fable 5.1Claude Fable 5Claude Opus 5GPT-5.6 Sol
Terminal-Bench-Science 0.152.6%24.7%29.0%22.4%
Terminal-Bench 4.055.8%42.0%52.3%37.3%
CursorBench 3.2.073.4%70.5%70.0%67.2%
AutomationBench31.4%17.1%26.9%19.6%
GDPval-AA v2 (Elo)1,8531,7231,8241,711
OSWorld 2.0 · partial77.9%72.9%75.4%Not reported
OSWorld 2.0 · strict41.7%36.1%39.6%Not reported
Humanity's Last Exam · no tools60.9%57.8%56.6%Not reported
Humanity's Last Exam · with tools65.0%63.8%63.6%Not reported

Artificial Analysis, independent

Top of the index, and it spends tokens to get there.

First of 192 models on the Intelligence Index, four points above Fable 5. It used about 70% more output tokens to get there, so each task costs more at the same token price.

Intelligence Index v4.1.1

Nine-evaluation composite

Agentic, coding, and reasoning evaluations combined into one score.

Index score

0 to 100

  • Claude Fable 5.166
  • Claude Fable 562
Artificial Analysis model pages for Fable 5.1 ↗ and Fable 5 ↗, read 1 Sep 2026.
View chart data
Artificial Analysis Intelligence Index values
ModelIndexRank
Claude Fable 5.166#1 of 192
Claude Fable 562#6 of 192

What it costs to run

Same price per token, more tokens per task

Cost per Intelligence Index task at Anthropic list prices.

Cost per task

$0 to $4

  • Claude Fable 5.13.69
  • Claude Fable 53.14
Output tokens, whole index
140Mvs 83M
Output speed
66 tok/svs 67
Artificial Analysis, own harness, Anthropic list prices. Up about 18% on Fable 5.
View chart data
Artificial Analysis cost and throughput values
MetricClaude Fable 5.1Claude Fable 5
Cost per Index task$3.69$3.14
Output tokens across the Index140M83M
Output speed66.0 tok/s66.9 tok/s
Time to first token285s122s

Pricing

List price unchanged. Cache reads cut 75%.

Anthropic's August usage data puts the saving at about 25% on typical workloads and up to 45% on cache-heavy agentic ones. Uncached, high-effort work costs more, as the per-task figure above shows.

How MiniRouter prices models →

Anthropic list, per 1M tokens

Input
$10.00unchanged
Output
$50.00unchanged
Cache write · 5 min
$12.50
Cache read
$0.25was $1.00

Anthropic's indexed workload cost

Fable 5 = 100

0 to 100

  • Fable 5, any workload100
  • Fable 5.1, typical75
  • Fable 5.1, agentic55

“About 25%” and “up to 45%” savings, plotted as 75 and 55.

List prices from the Fable 5.1 model docs ↗; savings claim from the announcement ↗.
View chart data
Anthropic list prices for Claude Fable 5.1
RateClaude Fable 5.1Claude Fable 5
Input / 1M$10.00$10.00
Output / 1M$50.00$50.00
Cache write, 5 min / 1M$12.50$12.50
Cache write, 1 hour / 1M$20.00$20.00
Cache read / 1M$0.25$1.00
Batch API50% off input and output50% off input and output

Before you switch from Fable 5

Three requests that used to work now return 400.

Anthropic's rules, enforced on every MiniRouter endpoint. Everything else carries over from Fable 5.

Anthropic migration guide ↗
  1. 01

    Forced tool use is rejected

    tool_choice "any" or "tool" returns 400. Use "auto" and name the tool in the prompt, or structured outputs for JSON.

  2. 02

    Thinking blocks are bound to the model

    Other models drop Fable 5.1 thinking blocks silently, so a mid-conversation switch loses that reasoning.

  3. 03

    Editing earlier turns invalidates thinking

    Accounts created from 31 Aug 2026 get a 400 when history before a thinking block changes. Keep harnesses append-only.

One route, standard clients.

Authenticate with a MiniRouter bearer key. Caller-provided upstream credentials are not used on this route.

Read model setup docs →
Base URLhttps://api.minirouter.sh/v1
Model IDanthropic/claude-fable-5.1
Endpoints
/v1/chat/completions/v1/messages/v1/responses
Parameters

max_tokens, temperature, stop, tools, tool_choice, reasoning, include_reasoning

Configured provider

Priority follows the live catalog order.

Input: text, image, pdf · Output: text

ProviderPriorityInput / 1MOutput / 1MParametersAccess
vercel
1$10.5$52.5max_tokens, temperature, stop, tools, tool_choice, reasoning, include_reasoningMiniRouter balance