Editorial illustration for Claude Opus 5 cheaper than Fable 5 but still trails on fact accuracy
Claude Opus 5 Cheaper Than Fable 5, Trails on Facts
Anthropic released Claude Opus 5 this week, and Artificial Analysis already ranks it as the most capable model on the market, edging out Claude Fable 5 while charging significantly less per token. The firm's Intelligence Index, a composite of nine tests spanning coding, scientific reasoning, and factual recall, puts Opus 5 at 61, just ahead of Fable 5's 60 and GPT-5.6 Sol's 59.
The catch shows up in how the model behaves under uncertainty. Opus 5 answers more often than it stays silent, even when it isn't sure, and that habit pushes its hallucination rate to 50 percent. For a model marketed on analytical strength and knowledge-based tasks, that's a real tension: score higher on benchmarks by attempting more answers, but risk more wrong ones along the way.
Anthropic worked directly with Artificial Analysis to test Opus 5 ahead of its public release, giving the numbers some weight before independent users get their hands on it. The pricing comparison against Fable 5 is where things get interesting, especially once you factor in performance at the "high" and "xhigh" settings against Anthropic's own earlier models.
Why this matters
For teams picking a model on price and benchmark scores alone, Opus 5 looks like an easy call: cheaper than Fable 5, ahead on the Intelligence Index at 61, strong on analytical and knowledge tasks. But that 50 percent hallucination rate, up 14 points because the model now answers more often instead of admitting uncertainty, is the number that should give developers pause before shipping it into anything customer-facing or research-adjacent. A model that talks more but knows the same amount isn't more useful, it's just more confident about being wrong.
Anthropic's 7-point gain on AA-Omniscience shows real progress on factual grounding, but "improved, still trails Fable 5" is not a claim you want to build a support bot or a research assistant on without guardrails. For founders weighing cost against risk, the math changes depending on the use case: a coding tool can tolerate more wrong guesses than a legal or medical assistant can. The lesson from Epoch AI's numbers is straightforward: benchmark leaderboards reward boldness, not honesty, and right now those two things are pulling in opposite directions.
Related Reading
Browse all Anthropic news →
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic Trump cracks down on Anthropic after Amazon tip; staff largely foreign
Open Source
Related Topic Anthropic says Claude writes >90% of its code; AI pause button urged
LLMs & Generative AI
From Archives South Korea Charts AI Future With NVIDIA at Summit
From Archives Instella-MoE Language Model Improves to 73.22 Score After Post-Training
From Archives Anthropic Launches Opus 5 AI Model, Completes Series Update
Business & Startups
From Archives Anthropic's Opus 5 AI Nears Fable 5 Capabilities, Excels at Coding
Open Source
Common Questions Answered
How does Claude Opus 5's pricing compare to Claude Fable 5?
Claude Opus 5 charges significantly less per token than Claude Fable 5, making it a more cost-effective option for teams evaluating models based on price. Despite the lower cost, Opus 5 still ranks ahead of Fable 5 on the Artificial Analysis Intelligence Index with a score of 61 compared to Fable 5's 60.
What is Claude Opus 5's score on the Artificial Analysis Intelligence Index?
Claude Opus 5 scored 61 on the Artificial Analysis Intelligence Index, which combines nine tests covering knowledge work, coding, scientific reasoning, and factual accuracy. This score puts it just ahead of Claude Fable 5 (60) and GPT-5.6 Sol (59), making it the most capable model on the market according to the ranking.
What is the main concern with Claude Opus 5's hallucination rate?
Claude Opus 5 has a 50 percent hallucination rate, which increased by 14 points because the model now answers questions more often instead of admitting uncertainty. This higher hallucination rate should give developers pause before deploying the model in customer-facing or research-adjacent applications where accuracy is critical.
Why does Claude Opus 5 have a higher hallucination rate than previous versions?
Claude Opus 5's increased hallucination rate stems from a design choice to answer more questions rather than staying silent when uncertain. While this makes the model more responsive, it comes at the cost of reduced factual accuracy compared to models that are more conservative about admitting uncertainty.
Further Reading
- Claude Fable 5 vs Claude Opus 4.8: Complete Comparison - LLM Stats
- Claude Fable 5 vs Claude Opus 4.8 - CodingFleet
- Claude Fable 5 Debut: vs Opus 4.8, Sonnet 4.6, GPT-5.5 - AI Crucible
- Claude Fable 5 vs Opus 4.8: The Mythos Hype Meets Reality - Tessl
- Anthropic's Claude Fable 5 sets a new bar for enterprise content intelligence - Box Blog
