Is Claude Too Verbose? A Paying User's Real-World Comparison With GPT-4

A power user who subscribes to both Claude and GPT-4's premium tiers (at $200/month each) recently posted a complaint on Reddit: Claude's responses are far too long, and even simple questions get wall-of-text answers. The post quickly sparked a lively debate in the AI community about how to balance "conciseness" with "completeness."

An Extreme Example: A One-Sentence Question Gets a 124-Word Essay

The user shared a telling example — when he asked "which button should I click," Claude responded with 124 words. Even more striking: for a 3,400-word task prompt, GPT-4 wrapped it up in just 81 words, while Claude left him thoroughly frustrated.

In an attempt to rein in Claude's verbosity, the user tried a number of approaches:

• Using the ELI5 (Explain Like I'm 5) technique
• Adjusting memory settings
• Testing different versions (Opus 4.8, Opus 5, Fable 5.1)

None of these methods could make Claude "consistently" give concise replies.

Where GPT-4 Has the Edge on Brevity

The user had particular praise for GPT-4's response style:

• Works out of the box: No special prompting needed for it to understand a user's preference for brevity
• Clean formatting: Responses are well-structured with no redundant information
• Task-focused: Stays on target rather than over-explaining

This comparison highlights a key differentiator in how large models are productized — the fine-tuning of user experience details. GPT-4 seems to have mastered the art of "knowing when to stop," while Claude leans toward providing more comprehensive — and potentially over-bloated — answers.

The Root Causes of AI Verbosity

Claude's tendency to ramble isn't an isolated quirk; it stems from several technical and product-level factors:

Differences in training objectives: Anthropic's training of Claude places heavy emphasis on "helpfulness" and "safety," which may cause the model to default to providing more context and disclaimers. OpenAI, by contrast, appears to have invested more optimization resources into conciseness.

Side effects of RLHF: During Reinforcement Learning from Human Feedback (RLHF) training, human evaluators tend to favor responses that appear "more detailed," which teaches the model to over-explain over time.

Product positioning: Claude is positioned as a "safe and reliable AI assistant," and in certain scenarios it trades conciseness for completeness and accuracy.

Real-World Impact on Business Use Cases

The user made clear that both products serve different purposes in his workflow and that he wants to keep using Claude. This suggests that despite the verbosity issue, Claude still holds irreplaceable advantages in reasoning, code quality, or specific domains.

For enterprise users, however, verbose AI output isn't just an annoyance — it creates real costs and efficiency losses:

• Reduced productivity: More time spent sifting through responses to find the useful parts
• Higher usage costs: Longer responses consume more tokens, driving up API call expenses
• Increased integration complexity: Automated pipelines require additional post-processing steps

Practical Tips for Getting More Concise Responses from Claude

While the user found existing approaches limited, the following strategies are still worth trying:

1. Optimize your system prompt: At the start of each conversation, explicitly request "brief replies, no more than X words"
2. Few-shot example guidance: Provide ideal concise response samples for the model to reference
3. Structured output constraints: Ask the model to respond in a format like JSON to limit free-form elaboration
4. Fine-tune API parameters: When using the API, experiment with the temperature and max_tokens parameters

From a product standpoint, Anthropic may want to consider adding a "concise mode" toggle, letting users freely switch between thoroughness and brevity. This would not only address a clear user pain point but also make Claude more adaptable across different use cases.

This case is a reminder that in the age of large language models, raw technical capability is just the baseline. The real competitive edge lies in translating that capability into a product experience that actually meets user expectations.