Kimi K2.7 Code versus Claude Fable 5: landing pages 94% cheaper

Together AI conducted an experiment comparing Kimi K2.7 Code and Claude Fable 5 from Anthropic on 12 landing pages, and published the results on June 17, 2026. According to the test results, Kimi K2.7 Code turned out to be approximately 94% cheaper than Fable 5 and nearly matched it in result quality on most pages. The publication author is Hassan El Mghari.

How the comparison was conducted

Both models were given the same prompts to create landing pages from different categories: a developer tool that converts SQL queries into charts, a landing page for a B2B SaaS startup with task management and team workflows, and a page for an art-deco-style retro bar with gold and emerald colors in the spirit of the 1920s. With the first simple prompt without additional context, both models produced pages that looked "recognizably AI-generated."

• Compared a total of 12 landing pages across both engines
• Kimi K2.7 Code proved on average approximately 16 times cheaper than Claude Fable 5
• Kimi K2.7 Code was approximately 8 times cheaper than Claude Opus 4.8
• Categories of prompts — SQL-to-chart tool, B2B SaaS, and art-deco retro bar
• All variants, including those generated by Opus 4.8, Fable 5, and Kimi K2.7 Code, were published on the OVSC website

What changed the test results

To improve the design quality above the baseline level, the Together AI team connected both models to their own MCP server with design references — it transmitted screenshots of well-designed landing pages, individual UI elements, and other visual samples to the models. According to the experiment authors, this MCP server with references turned out to be the most effective way to give models a stronger creative direction than just a text prompt.

After adding such context, the quality gap between Kimi K2.7 Code and Claude Fable 5 noticeably narrowed — while Kimi continued to remain several times cheaper in price.

What this means

Together AI's experiment shows that open models like Kimi K2.7 Code are no longer just a "cheap alternative" to closed flagships — with appropriate context (in this case through design-MCP) they deliver comparable quality solutions at a fraction of the cost, and the gap between open and closed models in tasks like interface generation continues to narrow faster than expected.

Frequently Asked Questions

How much cheaper is Kimi K2.7 Code than Claude Fable 5?

According to Together AI, on average Kimi K2.7 Code was approximately 16 times cheaper than Claude Fable 5 and approximately 8 times cheaper than Claude Opus 4.8 with comparable result quality.

What exactly was compared in the experiment?

The team compared 12 landing pages generated by Kimi K2.7 Code and Claude Fable 5 with identical prompts — from a developer tool and B2B SaaS service to a themed retro bar website, and posted all variants on the OVSC website together with breakdowns by cost, tokens, and generation time.

What helped the models deliver higher quality results?

A custom MCP server with design references — screenshots of successful landing pages and individual UI elements, which gave the models a stronger creative direction than just a text prompt.