Code Migration
Whether a model can reimplement a working program in another language and preserve its behaviour — the work of porting a tool to a new ecosystem or modernising legacy code, including COBOL to Java.
How it is run
Two task types: CLI migration between modern languages, and COBOL-to-Java modernisation. A migration is scored by the share of hidden behaviour tests the rebuilt program passes, and a separate anti-cheat check zeroes wrappers, copied reference artifacts and wrong-language submissions.
What it cannot tell you
Automated migration is far from turnkey: even the leader passes only about two-thirds of the hidden behaviour tests, so the ports mostly work but are not drop-in replacements. Serving speed can decide a score before capability does — Kimi K3 reaches 60.0% on COBOL-to-Java but 1.5% on CLI migration because its endpoint's latency exhausts the harness clock before it submits anything gradeable.
- Leader
- Claude Sonnet 5.569.83% · Vals AI's own list
- Top-ten spread
- 16.91%Leader minus tenth place, or last place in smaller fields.
- Median
- 29.95%of the 71 rows published
- No published result
- 9of 46 models on this site
Independent leaderboard · Vals AI
Top 71 of 71 runs by Vals AI, one harness.
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 5.5 | 69.83% |
| 2 | Gemini 4 Argon | 68.17% |
| 3 | GPT-6 Astra | 67.74% |
| 4 | Claude Opus 5.5 | 66.65% |
| 5 | GPT-6.1 Sol | 65.12% |
| 6 | Claude Opus 5 | 57.47% |
| 7 | GPT-6 Sol | 57.20% |
| 8 | Claude Fable 5 | 55.06% |
| 9 | Claude Fable 5.1 | 54.61% |
| 10 | GPT-5.6 Sol | 52.92% |
| 11 | GPT-5.6 Terra | 47.80% |
| 12 | Hunyuan Hy4 Preview | 47.43% |
| 13 | Muse Spark 1.3 Max | 47.41% |
| 14 | Claude Opus 4.8 | 47.25% |
| 15 | DeepSeek V4.1 Flash | 45.62% |
| 16 | GPT-5.5 | 45.16% |
| 17 | Grok 4.7 | 44.82% |
| 18 | Grok 4.6 | 44.57% |
| 19 | GPT-5.6 Luna | 44.55% |
| 20 | Claude Sonnet 5 | 44.39% |
| 21 | GLM 5.3 | 44.22% |
| 22 | Claude Opus 4.7 | 43.88% |
| 23 | MiMo V2.6 Pro | 43.01% |
| 24 | GPT-6 Luna | 42.55% |
| 25 | DeepSeek V4 Pro | 41.54% |
| 26 | MiMo-V2.6-Flash | 40.93% |
| 27 | Claude Sonnet 4 6 | 39.89% |
| 28 | Deepseek V4 Flash 0731 | 38.63% |
| 29 | Glm 5.2 | 37.87% |
| 30 | Grok 4.5 | 36.60% |
| 31 | Gemini 3.8 Flash | 36.55% |
| 32 | Gpt 5.4 2026 03 05 | 34.98% |
| 33 | Gemini 3.7 Flash | 34.80% |
| 34 | Muse Spark 1 1 | 31.11% |
| 35 | Gemini 3.6 Flash | 30.93% |
| 36 | Muse Spark 1.2 | 29.95% |
| 37 | Kimi K2.6 | 27.77% |
| 38 | Muse Spark 1 3 | 27.58% |
| 39 | Gemini 3.5 Flash | 26.75% |
| 40 | Deepseek V4 Pro | 26.20% |
| 41 | Glm 5.1 | 25.77% |
| 42 | Kimi K2.7 Code | 25.39% |
| 43 | Qwen3.8 Max | 23.96% |
| 44 | Mimo V2.5 Pro | 21.56% |
| 45 | GLM 5.3 Flash | 20.52% |
| 46 | MiniMax M3 | 19.93% |
| 47 | Gemini 3.1 Pro | 17.31% |
| 48 | Kimi K3 | 16.10% |
| 49 | Gpt 5.4 Nano 2026 03 17 | 14.47% |
| 50 | Mimo V2.5 | 14.23% |
| 51 | Qwen3.8 27b | 14.16% |
| 52 | Inkling Small | 13.68% |
| 53 | Qwen3.7 Max | 13.07% |
| 54 | Gpt 5.4 Mini 2026 03 17 | 12.94% |
| 55 | Qwen3.7 Plus | 12.86% |
| 56 | Inkling | 11.79% |
| 57 | Qwen3.6 Plus | 11.10% |
| 58 | MiniMax M2.7 | 8.69% |
| 59 | Claude Haiku 4 5 20251001 Thinking | 7.55% |
| 60 | Kimi K2.5 Thinking | 6.96% |
| 61 | Grok 4.3 | 6.79% |
| 62 | Gemini 3 Flash Preview | 6.39% |
| 63 | Gemini 3.5 Flash Lite | 6.15% |
| 64 | Mistral Medium 3.5 | 5.13% |
| 65 | Nemotron 3 Ultra 550b A55b | 4.91% |
| 66 | Gemini 3.1 Flash Lite Preview | 4.61% |
| 67 | Mercury 2.5 | 4.49% |
| 68 | Nemotron Lightning 3p5 30b A3b | 2.95% |
| 69 | Ling 3.0 Flash Af Rc3 | 1.82% |
| 70 | Grok 4.20 0309 Reasoning | 0.31% |
| 71 | Ling 3.0 Flash 2607 | 0.00% |
No published result tracked
9 of the 46 models on this site have no published number on this benchmark, so their cells in the home table stay blank rather than reading as zero.
Read this with
Benchmarks in the same category, which usually disagree with this one on purpose.
Sources
Every page used for this benchmark, read on 17 October 2026.
- Benchmark site or the page that publishes itwww.vals.ai/benchmarks/code-migration
- Independent resultswww.vals.ai/benchmarks/code-migration