BenchLeader
Menu
LeaderboardsModelsProvidersBenchmarksCompareReleases1BlogMethodology

Code Migration

Rewriting whole programs in another language so they still pass the tests. Run by Vals AI.

As of 19 Sept 2026, GPT-6 Astra leads Code Migration on BenchLeader with 67.7%, ahead of Claude Opus 5 at 57.5%, across 58 model configurations with a published result.

Published by
Vals AI
Category
Coding
Index weight
Reference only
Models
58
Data as of
19 Sept 2026

Vals AI (vals.ai).

What the test looks like

The model receives a working program and must port it to a different language; the port is checked against the original test suite.

How it is scored

Percent of migrations passing, run by Vals AI.

What to keep in mind

A specific engineering skill; strong scores here do not imply strong algorithmic coding.

Top 15
  1. 1GPT-6 Astra (max)67.7%
  2. 2Claude Opus 557.5%
  3. 3Claude Fable 555.1%
  4. 4Claude Fable 5.154.6%
  5. 5GPT-5.6 Sol (max)52.9%
  6. 6GPT-5.6 Terra (max)47.8%
  7. 7Muse Spark 1.3 (max)47.4%
  8. 8Claude Opus 4.847.3%
  9. 9DeepSeek V4.1 Flash (high)45.6%
  10. 10GPT-5.5 (xhigh)45.2%
  11. 11Grok 4.6 (high)44.6%
  12. 12GPT-5.6 Luna (max)44.5%
  13. 13Claude Sonnet 544.4%
  14. 14GLM-5.3 (max)44.2%
  15. 15Claude Opus 4.743.9%
58 of 58
#
1GPT-6 AstramaxOpenAI67.7%71.82026-09-16
2Claude Opus 5Anthropic57.5%66.72026-09-16
3Claude Fable 5Anthropic55.1%68.32026-09-16
4Claude Fable 5.1Anthropic54.6%64.52026-09-16
5GPT-5.6 SolmaxOpenAI52.9%68.82026-09-16
6GPT-5.6 TerramaxOpenAI47.8%65.02026-09-16
7Muse Spark 1.3maxMeta47.4%69.32026-09-16
8Claude Opus 4.8Anthropic47.3%61.92026-09-16
9DeepSeek V4.1 FlashhighDeepSeekopen ↗45.6%–2026-09-16
10GPT-5.5xhighOpenAI45.2%67.62026-09-16
11Grok 4.6highxAI44.6%64.52026-09-16
12GPT-5.6 LunamaxOpenAI44.5%60.22026-09-16
13Claude Sonnet 5Anthropic44.4%57.22026-09-16
14GLM-5.3maxZhipu AIopen ↗44.2%65.82026-09-16
15Claude Opus 4.7Anthropic43.9%64.52026-09-16
16DeepSeek V4 Pro 0813maxDeepSeekopen ↗41.5%56.02026-09-16
17Claude Sonnet 4.6Anthropic39.9%59.02026-09-16
18DeepSeek V4 Flash 0731highDeepSeekopen ↗38.6%55.02026-09-16
19GLM-5.2maxZhipu AIopen ↗37.9%63.92026-09-16
20Grok 4.5highxAI36.6%61.02026-09-16
21Gemini 3.8 FlashhighGoogle36.5%64.52026-09-16
22GPT-5.4xhighOpenAI35.0%65.42026-09-16
23Gemini 3.7 FlashhighGoogle34.8%64.62026-09-16
24Muse Spark 1.1xhighMeta31.1%61.82026-09-16
25Gemini 3.6 FlashhighGoogle30.9%61.72026-09-16
26Muse Spark 1.2xhighMeta29.9%64.12026-09-16
27Kimi K2.6Moonshot AIopen ↗27.8%60.52026-09-16
28Muse Spark 1.3xhighMeta27.6%68.02026-09-16
29Gemini 3.5 FlashhighGoogle26.8%63.62026-09-16
30DeepSeek V4 PromaxDeepSeekopen ↗26.2%64.22026-09-16
31GLM-5.1Zhipu AIopen ↗25.8%56.92026-09-16
32Kimi K2.7 CodeMoonshot AIopen ↗25.4%55.92026-09-16
33Qwen3 8maxAlibaba24.0%66.52026-09-16
34MiMo-V2.5-ProXiaomiopen ↗21.6%59.22026-09-16
35GLM-5.3-FlashmaxZhipu AIopen ↗20.5%55.32026-09-16
36MiniMax-M3MiniMaxopen ↗19.9%57.52026-09-16
37Gemini 3.1 ProhighGoogle17.3%60.72026-09-16
38Kimi K3maxMoonshot AIopen ↗16.1%67.22026-09-16
39GPT-5.4 nanohighOpenAI14.5%49.12026-09-16
40Qwen3.8 27BxhighAlibabaopen ↗14.2%59.22026-09-16
41Inkling SmallThinking Machinesopen ↗13.7%56.32026-09-16
42Qwen3 7maxAlibaba13.1%61.02026-09-16
43GPT-5.4 minixhighOpenAI12.9%55.72026-09-16
44Qwen3.7 PlusAlibaba12.9%59.92026-09-16
45Qwen3.6 PlusAlibabaopen11.1%58.52026-09-16
46MiniMax-M2.7MiniMaxopen ↗8.7%56.12026-09-16
47Claude Haiku 4.5thinkingAnthropic7.5%47.52026-09-16
48Kimi K2.5thinkingMoonshot AIopen7.0%58.52026-09-16
49Grok 4.3highxAI6.8%58.22026-09-16
50Gemini 3 FlashhighGoogle6.4%58.72026-09-16
51Gemini 3.5 Flash LitehighGoogle6.2%44.32026-09-16
52Mistral Medium 3.5highMistral AIopen5.1%–2026-09-16
53Nemotron 3 Ultra 550B A55BNVIDIAopen ↗4.9%48.12026-09-16
54Gemini 3.1 Flash LitehighGoogle4.6%49.12026-09-16
55Mercury 2.5highInception4.5%–2026-09-16
56Nemotron Lightning 3.5 30B A3bNVIDIA3.0%–2026-09-16
57Grok 4.20thinkingxAI0.3%59.92026-09-16
58Ling 3.0 FlashAnt Groupopen ↗0.0%53.62026-09-16

Cite as: BenchLeader, “Code Migration leaderboard”, https://www.benchleader.com/benchmarks/vals_code_migration, data as of 19 Sept 2026.

Code Migration: questions

What does Code Migration measure?
The model receives a working program and must port it to a different language; the port is checked against the original test suite. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads Code Migration?
GPT-6 Astra leads Code Migration with 67.7% as of 19 Sept 2026, ahead of Claude Opus 5 at 57.5%.
How many models have Code Migration results?
58 model configurations have a Code Migration result on BenchLeader, all taken from Vals AI.
Who runs Code Migration and how often is it updated?
Code Migration is published by Vals AI. BenchLeader re-reads the published results every morning and records the date each result was published.
Does Code Migration count toward the BenchLeader Index?
No. Code Migration is shown for reference but left out of the composite index.