Code Translation2020en

TransCoder Evaluation on GeeksForGeeks Algorithmic Problems

Code translation benchmark between C++, Java, and Python using 852 GeeksForGeeks programming problems. Each function has correct implementations in all three languages. Primary metric is Computational Accuracy (CA) — translated code must pass all unit tests.

Samples:852
Metrics:computational-accuracy
Paper / Website
Current State of the Art

Claude Sonnet 4

Anthropic

89.4

computational-accuracy

computational-accuracy Progress Over Time

Showing 5 breakthroughs from Jun 2020 to Mar 2026

66.672.879.085.391.5Jun 2020Nov 2021Apr 2023Oct 2024Mar 2026computational-accuracyDate

Key Milestones

Jun 2020
TransCoder

Java→Python translation. TransCoder paper Table 3. CA (% of functions passing tests).

68.7
Source
May 2023
CodeT5+

Java→Python translation. CodeT5+ paper (220M encoder-decoder variant).

72.1
+4.9%
Source
Feb 2024
StarCoder2-15B

Java→Python translation. StarCoder2 paper. Instruction-tuned model.

78.4
+8.7%
Source
Jun 2024
GPT-4o

Java→Python translation. Reported in DeepSeek-Coder-V2 comparative table.

88.2
+12.5%
Source
Mar 2026
Claude Sonnet 4Current SOTA

Java→Python translation. Anthropic model card.

89.4
+1.4%
Source
Total Improvement
30.1%
Time Span
5y 11m
Breakthroughs
5
Current SOTA
89.4

Top Models Performance Comparison

Top 7 models ranked by computational-accuracy

computational-accuracy1Claude Sonnet 489.4100.0%2GPT-4o88.298.7%3Qwen2.5-Coder-32B-Instruct86.396.5%4DeepSeek-Coder-V2-Instruct84.694.6%5StarCoder2-15B78.487.7%6CodeT5+72.180.6%7TransCoder68.776.8%0%25%50%75%100%% of best
Best Score
89.4
Top Model
Claude Sonnet 4
Models Compared
7
Score Range
20.7

computational-accuracyPrimary

#ModelScorePaper / CodeDate
1
Claude Sonnet 4API
Anthropic		89.4
official-model-card
Mar 2026
2
GPT-4oAPI
OpenAI		88.2
DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code IntelligenceCode
Jun 2024
3
Qwen2.5-Coder-32B-InstructOpen Source
Alibaba		86.3
Qwen2.5-Coder Technical ReportCode
Sep 2024
4
DeepSeek-Coder-V2-InstructOpen Source
DeepSeek		84.6
DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code IntelligenceCode
Jun 2024
5
StarCoder2-15BOpen Source
BigCode		78.4
StarCoder2 and The Stack v2: The Next GenerationCode
Feb 2024
6
CodeT5+Open Source
Salesforce		72.1
CodeT5+: Open Code Large Language Models for Code Understanding and GenerationCode
May 2023
7
TransCoderOpen Source
Meta		68.7
Unsupervised Translation of Programming LanguagesCode
Jun 2020

Related Papers5

Qwen2.5-Coder Technical Report
Sep 2024Models: Qwen2.5-Coder-32B-Instruct
PaperCode
DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence
Jun 2024Models: GPT-4o, DeepSeek-Coder-V2-Instruct
PaperCode
StarCoder2 and The Stack v2: The Next Generation
Feb 2024Models: StarCoder2-15B
PaperCode
CodeT5+: Open Code Large Language Models for Code Understanding and Generation
May 2023Models: CodeT5+
PaperCode
Unsupervised Translation of Programming Languages
Jun 2020Models: TransCoder
PaperCode