Brainstorming Prompts

For the log line prompt, both Claude Sonnet 4 and Claude Opus 4 produced reasonably good fantasy story ideas, with Opus 4 perhaps having a slight edge. The responses contained common fantasy tropes and cliches, but that is to be expected given the vague prompt. Overall, the quality seems on par with or slightly better than the previous 3.7 versions of the models.

Outlining Prompts

Both Sonnet 4 and Opus 4 were able to generate full 40-chapter outlines, with Opus 4 providing more detailed scene descriptions. The outlines contained some issues, such as the "save the cat" moments not quite hitting the mark, but the overall quality was good and on par with the previous 3.7 versions.

Prose Prompts

The prose samples from both models were decent, with Opus 4 providing slightly more polished and detailed descriptions. However, the differences were minor, and both models performed about as well as the 3.7 versions.

Dialogue Prompts

The dialogue generated by both Sonnet 4 and Opus 4 was acceptable, but did not show a significant improvement over the previous 3.7 versions.

Editing Prompts

Both models performed well on the editing prompts, with Opus 4 showing a more substantial improvement over the original scene compared to Sonnet 4.

Marketing Prompts

For the ad headline prompts, Opus 4 outperformed Sonnet 4, generating more concise and intriguing headlines.

SEO Article Prompts

Both models produced reasonably good SEO articles, though they fell short of the requested 4,000-word count.

Overall, the improvements in Claude 4 over the previous 3.7 versions are incremental rather than transformative. The models perform well and are suitable for most creative writing tasks, but the differences are not dramatic. Users of the previous 3.7 versions may not feel a strong need to upgrade, but those new to the Claude models will find the 4.0 versions to be capable and reliable tools.

In the latest release, OpenAI has announced the next generation of Claude models - Claude Opus 4 and Claude Sonnet 4. These models claim to set new standards for coding, advanced reasoning, and AI agents.

Claude Opus 4 is touted as the "world's best coding model" with sustained performance on complex, long-running tasks and agent workflows. Claude Sonnet 4, on the other hand, is a significant upgrade to the previous Claude Sonnet 3.7 model, promising superior coding and reasoning while responding more precisely to instructions.

While the focus of these models seems to be on coding and advanced reasoning, it will be interesting to see how they perform in creative writing tasks, which has been a strength of the Claude family of models.

In my initial testing, I found that both Sonnet 4 and Opus 4 performed reasonably well on the creative writing tasks, producing decent results that were on par with or slightly better than the previous 3.7 Sonnet model. The log line and outlining prompts yielded satisfactory responses, with Opus 4 perhaps having a slight edge in terms of the level of detail and coherence.

However, when it came to the prose writing and dialogue prompts, the improvements were more subtle. The writing still had a somewhat AI-generated feel, with occasional awkward phrasing and overuse of certain stylistic devices like em-dashes. Improving the prompt did seem to yield better results, suggesting that these models still require careful prompting to bring out their full potential.

The editing prompt, on the other hand, showed more significant improvements, with both Sonnet 4 and Opus 4 demonstrating a better ability to refine and enhance the provided text compared to previous models.

Overall, while the Claude 4 models do appear to be an incremental improvement over their predecessors, the leap in creative writing performance is not as dramatic as the claims around coding and reasoning. For authors and writers, the Sonnet 4 model may be the more cost-effective option, as it seems to perform on par with the more expensive Opus 4 in the creative writing tasks I tested.

As with any AI model, it's important to thoroughly test and evaluate these tools for your specific needs and use cases. The Claude family of models continues to be a strong contender in the creative writing space, but the latest iterations may not represent a revolutionary leap forward just yet. Ongoing refinement and iteration will likely be necessary to truly unlock their full potential.

When it comes to the brainstorming prompt, both Claude for Sonnet and Claude for Opus performed reasonably well, producing a variety of fantasy log line ideas.

For the Sonnet model, the log lines included common fantasy tropes and cliches, but were still coherent and imaginative. The quality was on par with, or slightly better than, the previous 3.7 version of Sonnet.

The Opus model, on the other hand, generated log lines that were slightly more unique and compelling. While still drawing from familiar fantasy elements, the ideas felt a bit more original and intriguing.

Overall, both models demonstrated solid brainstorming capabilities, with Opus potentially having a slight edge in terms of creativity and inventiveness. However, the differences were relatively minor, and users comfortable with the previous 3.7 Sonnet model should find the 4.0 version to be a reliable tool for generating fantasy story ideas.

When it comes to the outlining and scene building capabilities of Claude 4 Sonnet and Claude 4 Opus, the results are quite promising.

Outlining

• Both Sonnet 4 and Opus 4 were able to provide detailed, multi-chapter outlines, with Opus 4 even generating an 11,000-word outline.
• The outlines contained solid plot points, character introductions, and dramatic moments, showcasing the models' ability to construct cohesive narrative structures.
• While there were a few instances where the models struggled to fully capture the "save the cat" moment or transition between chapters as seamlessly as desired, the overall quality of the outlines was quite good.
• Compared to previous versions of the models, the outlining capabilities appear to have improved, with the ability to generate longer, more comprehensive outlines.

Scene Building

• When it came to fleshing out specific scenes, both Sonnet 4 and Opus 4 provided detailed descriptions, incorporating sensory details and character actions.
• The prose, while not entirely free of some AI-esque phrasing, was generally well-written and evocative, capturing the tone and mood of the scenes.
• Interestingly, Opus 4 tended to provide slightly more thorough and nuanced scene descriptions compared to Sonnet 4, suggesting a potential edge in this area.
• The models also demonstrated the ability to improve the quality of the prose when provided with more detailed prompts, indicating their responsiveness to more specific instructions.

Overall, the outlining and scene building capabilities of Claude 4 Sonnet and Claude 4 Opus appear to be solid, with noticeable improvements over previous versions of the models. While not perfect, the models show promise in their ability to assist writers in the early stages of the creative process, from brainstorming to outlining and scene development.

When it comes to the dialogue prompt, both Claude for Sonnet and Claude for Opus performed similarly. The dialogue felt natural and in-character, with some good back-and-forth between the characters. There weren't any major issues with the flow or authenticity of the exchanges.

For the editing prompt, where the AI was given a subpar version of a scene and asked to improve it, both models did a solid job. The edited versions were more polished, with tighter writing and fewer overly dramatic phrases. Opus in particular seemed to do a slightly better job at refining the scene while staying true to the original intent.

Overall, the dialogue and editing capabilities of these new Claude models appear to be on par with or slightly better than the previous 3.7 versions. They are able to craft natural-sounding conversations and make meaningful improvements to existing prose. This suggests they could be useful tools for authors looking to refine their dialogue and tighten up their writing.

Marketing Headline Samples

Sonnet 4:

• Prince discovers his entire life was a lie, but which betrayal will destroy him first?
• What really happened the day the King's Son was cast into the wasteland to die? Asking a lot of questions could get the rebels killed.
• Betrayed by blood, hunted by monsters, but the Zolith aren't the deadliest threat in the wasteland.
• Why would a father condemn his own son to face the Hungry alone? He thought the mountain city was sanctuary until he learned what they were really hiding.

Opus 4:

• King's Son was thrown from paradise, what he discovered outside changed everything.
• Three races ruled the wasteland, the fourth one shouldn't exist.
• They told him nothing survived beyond the mountains, they lied.
• A betrayed prince, a plague of the Hungry, a truth that could destroy a kingdom.
• He spent 20 years safe inside a mountain city, one day outside revealed 20 years of lies.
• When the king exiles his own son, the real question is what is he afraid of?
• The Hungry feast on flesh, but something else feasts on souls.

Overall, Opus 4 seems to produce more concise and intriguing marketing headlines compared to Sonnet 4. The Opus 4 headlines are more likely to grab a reader's attention.

SEO Article Performance

Sonnet 4: 2350 words (requested 4000) Opus 4: 2090 words (requested 4000)

Both models fell short of the requested 4000 word count for the SEO article prompt. The content produced was reasonably comprehensive, but lacked the depth and detail that would be expected for a full SEO article.

In terms of quality, the writing was acceptable but not exceptional. The Opus 4 model tended to provide more of an outline structure rather than fleshed out content. Overall, the SEO article performance was adequate but not outstanding for either model.

Based on the transcript provided, here is a concise summary of the key points regarding the performance of Claude 4 Sonnet and Claude 4 Opus:

• The author found the performance of Claude 4 Sonnet and Claude 4 Opus to be generally on par with or slightly better than the previous 3.7 versions, but not a major leap forward.

• For brainstorming and outlining tasks, Opus seemed to have a slight edge over Sonnet, providing more detailed and cohesive responses. However, the differences were relatively minor.

• In terms of dialogue and prose writing, the models performed comparably to 3.7, with some room for improvement in areas like adhering to instructions and avoiding overly dramatic phrasing.

• The author noted that the models may perform differently in other genres beyond fantasy, and recommended testing them thoroughly for specific use cases.

• Overall, the author concluded that while Claude 4 represents an incremental improvement, the differences from 3.7 are not dramatic enough to warrant an immediate switch for most users. Continued testing and iteration will likely lead to more substantial improvements in future versions.