✦ Our pick · Change existing / legacy code
Claude Opus 5.5 at medium effort, $17.97 per finished task, is the cheapest you can get today. Within our measurement error it ties with Claude Sonnet 5.5: equally good bets, worth trying two.
Updated Oct 13, 2026, every 3 hours. See it live with your plans and costs.
The top 10 for change existing / legacy code
Adapt, refactor or migrate a large existing codebase. Cost per finished task: the tokens at each model's best effort level for this work, plus your time fixing failed attempts ($93 an hour). Success is the share of such tasks a model gets right the first time, on the benchmarks below.
| # | Model | Effort | Success | $/task |
|---|---|---|---|---|
| 1 | Gemini 4 Argon Google | high | 68.2% partly measured | $16.92 |
| 2 | Claude Opus 5.5 Anthropic | medium | 67.3% partly measured | $17.97 |
| 3 | Claude Sonnet 5.5 Anthropic | xhigh | 66.0% partly measured | $18.98 |
| 4 | GPT-6.1 Sol OpenAI | medium | 61.4% partly measured | $21.01 |
| 5 | GPT-6 Astra OpenAI | low | 65.7% measured | $23.04 |
| 6 | Claude Opus 5 Anthropic | high | 63.5% partly measured | $24.42 |
| 7 | GPT-6 Sol OpenAI | max | 59.3% partly measured | $26.92 |
| 8 | Muse Wonder Meta | max | 62.5% partly measured | $28.74 |
| 9 | Claude Fable 5.1 Anthropic | high | 58.1% measured | $30.76 |
| 10 | GPT-5.6 Sol OpenAI | medium | 52.2% partly measured | $32.64 |
Best within your tool
| Tool | Model | Effort | $/task |
|---|---|---|---|
| Cursor | Claude Opus 5.5 | medium | $17.97 |
| Claude Code | Claude Opus 5.5 | medium | $17.97 |
| Codex | GPT-6.1 Sol | medium | $21.01 |
| Antigravity | Claude Opus 5.5 | medium | $17.97 |
| GitHub Copilot | Claude Opus 5.5 | medium | $17.97 |
If tokens come with your plan
Tokens included in a plan cost you nothing extra, so the pick is the most reliable model in it, at the leanest effort level within our error of the best (the top levels only use up the plan sooner).
| Plan | Model | Effort |
|---|---|---|
| Claude Pro | Claude Opus 5.5 | high |
| Claude Max | Claude Opus 5.5 | high |
| ChatGPT Plus or Business | GPT-6 Astra | low |
| ChatGPT Pro | GPT-6 Astra | low |
| Google AI Pro | Claude Opus 5.5 | high |
| Google AI Ultra | Claude Opus 5.5 | high |
Measured on
- Code Migration · Vals AI. Rewrite 30 working open-source programs into other languages (Python, Java, Kotlin, Rust, C++) and COBOL into Java; scored by hidden behaviour tests. Measured for 37 of the 42 ranked models, results as of Oct 12, 2026.
- FrontierCode · Cognition (via Epoch AI). Whether a model’s change to a real codebase could be merged without human edits. Measured for 26 of the 42 ranked models, results as of Oct 13, 2026.
- SWE Atlas Refactoring · Scale AI. 70 tasks: restructure real code (decompose, extract, move, evolve interfaces) while keeping its behaviour, graded on maintainability and cleanup too. Measured for 8 of the 42 ranked models.
How every number is worked out, with the errors of our estimates: the method.