{"slug": "same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex", "title": "Same code, about 60% more output tokens on Claude: 80 runs across Claude Code, Codex and Antigravity in 5 languages", "summary": "A developer ran 80 tests across Claude Code, Codex and Google's Antigravity CLI in five prompt languages and found that Claude Opus 5.5 consumed roughly 60% more output tokens per character of generated code than Gemini 4 Argon and about 57% more than gpt-6-luna. Because Opus scored nearly identically in two different tools (591 and 596 tokens per character), the developer attributes the gap to the model's tokenizer rather than the surrounding tool, noting that output tokens cost five times more than input on the API. An earlier test found the same source code in 10 programming languages needed 43% to 59% more Claude tokens than GPT tokens.", "body_md": "Yesterday I compared Google's new Antigravity CLI with Claude Code and Codex. One number kept bothering me: Claude Opus 5.5 wrote shorter games than Gemini but did not use proportionally fewer output tokens. So I divided output tokens by the size of the file each run wrote, across four tool and model pairs and five prompt languages.\n\n| Tool + model | EN | KO | JA | ES | UK | Average | \n|---|---|---|---|---|---|---|\n| Claude Code + Opus 5.5 | 554 | 624 | 581 | 614 | 584 | 591 | \n| Antigravity + Opus 5.5 | 574 | 590 | 606 | 612 | 597 | 596 | \n| Antigravity + Gemini 4 Argon | 367 | 380 | 370 | 364 | 378 | 372 | \n| Codex + gpt-6-luna | 367 | 386 | 385 | 366 | 376 | 376 | \n\nOpus needed about 60% more output tokens per character than Gemini, and about 57% more than GPT. The interesting part is that Opus came out almost the same in two very different tools (591 and 596). That points at the model's tokenizer, not the tool around it. The language of the prompt barely mattered.\n\nOutput tokens are the expensive ones. On the API, Opus 5.5 charges five times more for output than for input, so this gap goes straight into the bill, on top of any difference in how much code each model decides to write.\n\nThis also fits an earlier test of mine, where the same source code in 10 programming languages needed 43% to 59% more Claude tokens than GPT tokens.\n\n| Input tokens | EN | KO | JA | ES | UK | \n|---|---|---|---|---|---|\n| Claude Code + Opus 5.5 | 96,462* | 78,638 | 79,266 | 79,030 | 79,787 | \n| Codex + gpt-6-luna | 51,674 | 48,538 | 48,660 | 48,201 | 48,816 | \n| Antigravity + Gemini 4 Argon | 13,041 | 13,289 | 12,408 | 12,629 | 13,447 | \n| Antigravity + Opus 5.5 | 9,822 | 11,316 | 9,983 | 9,945 | 10,958 | \n\n*Two of the five English Claude Code runs took an extra step.\n\nThe prompt is a few dozen tokens; the rest of every request is the tool's own instructions. Which tool you use decides the input size, not which language you write in.\n\nOne more detail: in Antigravity, Opus wrote longer games for Korean and Ukrainian prompts (about 31% and 20% more characters than in English), so its total output rose there. Per character it stayed the same.\n\n79 of 80 did, and every game showed its text in the prompt language. The one failure is a Ukrainian Codex game from the earlier test.", "url": "https://wpnews.pro/news/same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex", "canonical_source": "https://dev.to/jaehyun_cho_0dff271e0d2e5/same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex-and-2a7k", "published_at": "2026-10-23 03:05:51+00:00", "updated_at": "2026-10-23 03:11:41.246806+00:00", "lang": "en", "topics": ["large-language-models", "ai-tools", "ai-agents", "developer-tools"], "entities": ["Claude Opus 5.5", "Claude Code", "Codex", "Antigravity", "Gemini 4 Argon", "gpt-6-luna", "Google", "Anthropic"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex", "markdown": "https://wpnews.pro/news/same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex.md", "text": "https://wpnews.pro/news/same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex.txt", "jsonld": "https://wpnews.pro/news/same-code-about-60-more-output-tokens-on-claude-80-runs-across-claude-code-codex.jsonld"}}