Same TypeScript Costs 73% More on Claude Than GPT: Tokenizer Differences Ignored
gabrielchua · x · 2026-08-17
Playcode blog publishes an AI model token cost benchmark analysis, showing that different vendors' tokenizers split the same file into different token counts, making $/Mtok incomparable across vendors. Anthropic's newest tokenizer (Sonnet 5, Opus 4.8, Fable 5) produces 30% more tokens from the same code than their previous one, with list prices unchanged. On identical files, it produces 1.36-1.73x GPT's token count, with TypeScript worst at 1.73x. Effective prices: Opus 4.8's $5/$25 behaves like $7.50/$37.50, and Sonnet 5 after intro window like $4.50/$22.50. Measurement covers input tokenization only; output verbosity, thinking, and caching further affect total costs.
Related event: OpenAI Tokenizer 30% More Efficient Than Anthropic's(2 posts)→
More from Models
- GLM-5.3 review: Cleaner code and consistent long-horizon performance — khademinori · 2026-08-17
- Test: Qwen 3.8 27B Outperforms GPT-5.6 Sol in Complex SVG Animation Tasks — Lirezh · 2026-08-17
- Seedance 2.5 Tops Multi-Image-to-Video Benchmark with Elo 1400 — rohanpaul_ai · 2026-08-17
- Qwen2.5 2B runs on phones with 1GB RAM; MMLU jumps to 54.8 — alexcovo_eth · 2026-08-17
- Qwen3.8 Distilled 9B/4B/2B Released: MMLU Scores Double — alejandroll10 · 2026-08-17
- New Claude/GPT APIs reportedly force temperature to 1 — andersonbcdefg · 2026-08-17