The real prices of frontier models
playcode.io
July 13, 2026
11 min read
🔥🔥🔥🔥🔥
54/100
Summary
Anthropic's newest tokenizer, Sonnet 5, generates approximately 30% more tokens from the same code compared to its previous version. For identical files, it produces 1.36 to 1.73 times the token count of GPT, with TypeScript resulting in the highest increase at 1.73 times.
Key Takeaways
- Anthropic's new tokenizer (Sonnet 5, Opus 4.8, Fable 5) generates approximately 30% more tokens from the same code compared to its previous tokenizer without changing the list price.
- On identical files, Claude's tokenizer produces 1.36 to 1.73 times the token count of GPT's tokenizer, with TypeScript resulting in the highest increase at 1.73 times.
- The cost of using a model is determined by the number of tokens generated and the price per token, which varies significantly between different vendors' tokenizers.
- The effective pricing for Anthropic's Opus 4.8 is equivalent to $7.50 for $5.00 / $25.00 due to the increased token generation.
Community Sentiment
Positives
- Fable shines at creative tasks, but its sky-high cost makes it hard to justify for everyday use — Opus 4.6/4.7 is a more practical choice for many.
- OpenAI's documentation and tokenizer improvements are a significant step up from Anthropic, making it easier for developers to work with their models.
- Fable's unique capabilities can tackle obscure problems that other models struggle with, proving its value in niche scenarios.
Concerns
- Fable's pricing is so high it feels unaffordable for regular use, limiting its accessibility despite its impressive capabilities.
- Sonnet 5's performance is lackluster compared to Opus 4.8, making it seem almost useless for coding tasks — a poor price/performance ratio.
- Anthropic's tokenizer is considered inefficient, essentially charging users for extra tokens that don't necessarily enhance performance.
