Oleksandr Kuzmenko, EditorPublished 7 min readAI-assisted · editor-reviewed
TL;DR
- Scored 65 on the Artificial Analysis Intelligence Index v5.0.1 versus a median of 40.
- Extremely slow execution at 44.4 minutes per task with a TTFT of 235.06 seconds.
- Heavy verbosity (275M tokens generated) pushes the average task evaluation cost to $26.54.
Key facts
- Artificial Analysis Intelligence Index
- 65 (median: 40)
- Input Token Price
- $8.00 / 1M tokens
- Output Token Price
- $40.00 / 1M tokens
- Context Window
- 1,000,000 tokens
- Time to First Token (TTFT)
- 235.06s
- Average Task Duration
- 44.4 minutes (median: 12.8 min)
Source: Key facts — Artificial Analysis Intelligence Index, Context Window, Time to First Token (TTFT)
Source: Key facts. Artificial Analysis Intelligence Index: 65 (median: 40). Context Window: 1,000,000 tokens. Time to First Token (TTFT): 235.06s. Each figure is copied from the story.
- Artificial Analysis Intelligence Index65 (median: 40)
- Context Window1,000,000 tokens
- Time to First Token (TTFT)235.06s
Source: Key facts — Input Token Price, Output Token Price
Source: Key facts. Input Token Price: $8.00 / 1M tokens. Output Token Price: $40.00 / 1M tokens. The chart compares only values that share a unit. Each figure is copied from the story.
Metric chart
- $8.00 / 1M tokensInput Token Price
- $40.00 / 1M tokensOutput Token Price
- Input Token Price$8.00 / 1M tokens
- Output Token Price$40.00 / 1M tokens
High Intelligence Index Performance
According to independent benchmark data from Artificial Analysis, Anthropic's Claude Fable 5.2 (Max, Default Fallback) scored 65 on the Intelligence Index v5.0.1, placing it well above the class median of 40 and amongst leading models. The index aggregates nine evaluations, including AA-Briefcase v1.1, AutomationBench-AA, Terminal-Bench 4.0, FrontierCode 2.0, AA-Omniscience v1.1, GDP.pdf, HLE-Diamond, CritPt v2 (Hard Subset), and Terminal-Bench-Science 0.1. The model features a 1M token context window (equivalent to roughly 1,500 A4 pages) and accepts both text and image modalities.
Critical Latency and Throughput Constraints
While the model ranks among the leading models in intelligence, operational speed metrics are notably slow:
- Time to First Token (TTFT): 235.06 seconds
- Output Speed: 103 tokens per second
- Time per Intelligence Task: 44.4 minutes (compared to a category median of 12.8 minutes)
These delays mean that pipelines must be configured with extended client-side socket timeouts to avoid dropping connections before reasoning output begins streaming.
Pricing and Token Verbosity Economics
Fable 5.2 sits at the high end of proprietary API pricing:
- Input Tokens: $8.00 per 1M tokens (category median: $2.00)
- Output Tokens: $40.00 per 1M tokens (category median: $10.00)
- Weighted Task Cost: $26.54 per Intelligence Index task
During evaluation, Fable 5.2 generated 275M tokens—nearly four times the median volume of 72.8M tokens for comparable models. Teams deploying this model must enforce explicit output token bounds to prevent runaway expenditure during autonomous reasoning loops.
When to use
- Asynchronous batch processing of deep scientific and agentic tasks.
- Complex reasoning workflows where response latency is secondary to depth of analysis.
When NOT to use
- Real-time user-facing chatbots requiring sub-second latency.
- Cost-sensitive tasks with strict budget ceilings per API call.
What to do today
- Configure API timeouts to at least 300 seconds when calling Claude Fable 5.2.
- Implement strict max_tokens constraints to control costs driven by excessive verbosity.
Step names are taken from What to do today.
Step names are taken from What to do today. Configure API timeouts to at least 300 seconds when calling Claude Fable 5.2.. Implement strict max_tokens constraints to control costs driven by excessive verbosity.
- Configure API timeouts to at least 300 seconds when calling Claude Fable 5.2.
- Implement strict max_tokens constraints to control costs driven by excessive verbosity.