Browse › Natural Language Processing › Code Generation › MBPP
MBPP Benchmark (Code Generation)
Code Generation is an important field to predict explicit code or program structure from multimodal data sources such as incomplete code, programs in another programming language, natural language descriptions or execution examples. Code Generation tools can assist the development of automatic programming tools to improve programming productivity.
Source: Deep Learning for Source Code Modeling and Generation
Image source: Measuring Coding Challenge Competence With APPS
The archive carries no text for this table; the description above is the archive's text for the task Code Generation. archive 2025-07-28
Over time archive 2025-07-28
The chart needs JavaScript; the table below carries every value.
Direction inferred from the metric name, not from the archive: Accuracy (higher is better). Points are placed at the row's paper date; 98 of 99 rows carry one.
Results archive 2025-07-28
Archive rows end at the archive snapshot, 2025-07-28: no result published after that date is in this table. Rank is the archive's row order at that snapshot; not re-ranked here. Metric values are the archive's strings. Column headers sort the table in your browser; each row keeps its archive rank.
| Paper | Code | Ran Syntology | Report | |||||
|---|---|---|---|---|---|---|---|---|
| 1 | EG-CFG (DeepSeek-V3-0324) | 96.6 | – | Paper | Code | 2025 | linked, not harvested | report |
| 2 | QualityFlow (Sonnet-3.5) | 94.2 | – | Paper | – | 2025 | no code linked | report |
| 3 | o1-mini + MapCoder (Hamming.ai) | 93.2 | ✓ | Paper | Code | 2024 | 0 of 10 ran · 10 unverified | report |
| 4 | MGDebugger (DeepSeek-V3-0324) | 92.4 | – | Paper | Code | 2024 | 8 of 9 ran · 1 unverified | report |
| 5 | GPT-4 + AgentCoder | 91.8 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 6 | CodeSim (GPT4o) | 90.7 | – | Paper | Code | 2025 | 0 of 3 ran · 3 unverified | report |
| 7 | Jiutian-大模型 | 90.0 | – | – | – | not matched | report | |
| 8 | GPT-3.5 Turbo (ChatGPT) + AgentCoder | 89.9 | – | Paper | Code | 2023 | 1 of 1 ran · 0 unverified | report |
| 9 | MapCoder (GPT-4o) | 89.7 | – | Paper | Code | 2024 | 0 of 10 ran · 10 unverified | report |
| 10 | GPT-4 (ChatGPT Plus) | 87.5 | – | Paper | – | 2023 | no code linked | report |
| 11 | Claude 3 Opus | 86.4 | – | Paper | – | 2024 | no code linked | report |
| 12 | LPW (GPT-4o) | 84.8 | – | Paper | Code | 2024 | 0 of 9 ran · 9 unverified | report |
| 13 | GPT-3.5 Turbo + FlowGenScrum + Test | 83.8±0.6 | – | Paper | – | 2024 | no code linked | report |
| 14 | AFlow(GPT-4o-mini) | 83.4 | – | Paper | Code | 2024 | linked, not harvested | report |
| 15 | GPT-3.5 Turbo (ChatGPT) | 83.2 | – | Paper | – | 2023 | no code linked | report |
| 16 | EG-CFG (DeepSeek Coder 1.3b Instruct) | 83.2 | – | Paper | Code | 2025 | linked, not harvested | report |
| 17 | MapCoder (GPT-4) | 83.1 | – | Paper | Code | 2024 | 0 of 10 ran · 10 unverified | report |
| 18 | o1-mini + Language Agent Tree Search (Hamming.ai) | 82.3 | – | Paper | Code | 2023 | 7 of 9 ran · 2 unverified | report |
| 19 | GPT-4 (Bing Chat) | 82 | – | Paper | – | 2023 | no code linked | report |
| 20 | GPT-3.5 Turbo + Language Agent Tree Search | 81.1 | – | Paper | Code | 2023 | 7 of 9 ran · 2 unverified | report |
| 21 | MGDebugger (CodeQwen1.5) | 80.8 | – | Paper | Code | 2024 | 8 of 9 ran · 1 unverified | report |
| 22 | Claude 3 Haiku | 80.4 | – | Paper | – | 2024 | no code linked | report |
| 23 | GPT-4 (Self-Debugging with unit tests + trace) | 80.2 | – | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 24 | GPT-4 (few-shot) | 80 | ✓ | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 25 | Claude 3 Sonnet | 79.4 | – | Paper | – | 2024 | no code linked | report |
| 26 | Bard (PaLM 2/chat-bison-001) | 76.2 | – | Paper | – | 2023 | no code linked | report |
| 27 | GPT-3.5 Turbo (Self-Debugging with unit tests + trace) | 72.8 | – | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 28 | Claude | 71.4 | – | Paper | – | 2023 | no code linked | report |
| 29 | code-davinci-002 175B (Self-Debugging with unit tests + trace) | 70.8 | – | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 30 | GPT-3.5 Turbo (few-shot) | 70.8 | ✓ | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 31 | DeepSeek-Coder-Instruct 33B (few-shot) | 70 | – | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 32 | GPT-3.5 Turbo + INTERVENOR | 69.8 | – | Paper | Code | 2023 | linked, not harvested | report |
| 33 | code-davinci-002 175B + LEVER | 68.9 | – | Paper | Code | 2023 | 6 of 22 ran · 16 unverified | report |
| 34 | code-davinci-002 175B + CodeT | 67.7 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 35 | GPT-3.5 Turbo (3-shot) | 67.6 | ✓ | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 36 | code-davinci-002 175B + Reviewer | 66.9 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 37 | code-davinci-002 175B + Coder-Reviewer | 66.4 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 38 | StarCoder2-15B | 66.2 | – | Paper | Code | 2024 | linked, not harvested | report |
| 39 | DeepSeek-Coder-Base 33B (few-shot) | 66 | – | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 40 | Code Llama - Python 70B (3-shot) | 65.5 | ✓ | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 41 | DeepSeek-Coder-Instruct 6.7B (few-shot) | 65.4 | – | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 42 | code-davinci-002 175B + MBR-Exec | 63 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 43 | Code Llama 70B (3-shot) | 62.4 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 44 | Code Llama - Instruct 70B (3-shot) | 62.2 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 45 | code-davinci-001 175B + CodeT | 61.9 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 46 | code-davinci-002 175B (3-shot) | 61.4 | ✓ | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 47 | Unnatural Code Llama 34B (3-shot) | 61.2 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 48 | Mixtral 8x7B (3-shot) | 60.7 | – | Paper | Code | 2024 | 5 of 5 ran · 0 unverified | report |
| 49 | DeepSeek-Coder-Base 6.7B (few-shot) | 60.6 | – | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 50 | code-davinci-001 175B + MBR-Exec | 58.2 | – | Paper | Code | 2022 | 6 of 12 ran · 6 unverified | report |
| 51 | Code Llama - Instruct 34B (3-shot) | 57 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 52 | Code Llama - Python 34B (3-shot) | 56.2 | ✓ | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 53 | code-cushman-001 12B (CodeT) | 55.4 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 54 | Code Llama 34B (3-shot) | 55 | ✓ | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 55 | StarCoder 15.5B (Self-Debugging with unit tests + trace) | 53.2 | – | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 56 | StarCoder 15.5B | 52.7 | – | Paper | Code | 2023 | 2 of 2 ran · 0 unverified | report |
| 57 | GPT-3.5 Turbo | 52.2 | ✓ | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 58 | WizardCoder 15B | 51.8 | ✓ | Paper | Code | 2023 | 3 of 7 ran · 4 unverified | report |
| 59 | PaLM 2-S* (few-shot) | 50 | – | Paper | Code | 2023 | linked, not harvested | report |
| 60 | CodeGen-Mono 16B + CodeT | 49.5 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 61 | Code Llama - Instruct 13B (3-shot) | 49.4 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 62 | DeepSeek-Coder-Instruct 1.3B (few-shot) | 49.4 | – | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 63 | StarCoderBase 15.5B | 49 | – | Paper | Code | 2023 | 2 of 2 ran · 0 unverified | report |
| 64 | Code Llama - Python 13B (3-shot) | 49 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 65 | Qwen2idae-16x14B (4-shot) | 48.6 | – | Paper | Code | 2024 | linked, not harvested | report |
| 66 | code-cushman-001 12B + MBR-Exec | 48.3 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 67 | Code Llama - Python 7B (3-shot) | 47.6 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 68 | Mistral 7B (3-shot) | 47.5 | – | Paper | Code | 2023 | 9 of 11 ran · 2 unverified | report |
| 69 | CodeGen 16B + MBR-Exec | 47.3 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 70 | StarCoder 15.5B (3-shot) | 47.2 | – | Paper | Code | 2023 | 0 of 1 ran · 1 unverified | report |
| 71 | PaLM Coder 540B | 47 | – | Paper | Code | 2022 | 30 of 37 ran · 7 unverified | report |
| 72 | Code Llama 13B (3-shot) | 47 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 73 | CodeGen 16B + Coder-Reviewer | 46.2 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 74 | DeepSeek-Coder-Base 1.3B (few-shot) | 46.2 | – | Paper | Code | 2024 | 9 of 10 ran · 1 unverified | report |
| 75 | GPT-3.5 Turbo (few-shot) | 45.4 | – | Paper | Code | 2023 | linked, not harvested | report |
| 76 | Llama 2 70B (zero-shot) | 45 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 77 | Code Llama - Instruct 7B (3-shot) | 44.4 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 78 | CodeGen 16B + Reviewer | 44.1 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 79 | phi-1.5-web 1.3B | 43.5 | – | Paper | Code | 2023 | linked, not harvested | report |
| 80 | Branch-Train-Merge 4x7B (top-2) | 42.6 | – | Paper | Code | 2024 | linked, not harvested | report |
| 81 | Code Llama 7B (3-shot) | 41.4 | – | Paper | Code | 2023 | 3 of 5 ran · 2 unverified | report |
| 82 | Camelidae-8×34B (4-shot) | 41.4 | – | Paper | Code | 2024 | linked, not harvested | report |
| 83 | GPT-3.5 Turbo (0-shot) | 39.8 | – | Paper | Code | 2023 | linked, not harvested | report |
| 84 | Branch-Train-MiX 4x7B (sampling top-2 experts) | 39.4 | – | Paper | Code | 2024 | linked, not harvested | report |
| 85 | LLaMA 65B (0-shot) | 37.7 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 86 | PaLM 540B | 36.8 | – | Paper | Code | 2022 | 30 of 37 ran · 7 unverified | report |
| 87 | SantaCoder 1.1B | 35 | – | Paper | Code | 2023 | 2 of 2 ran · 0 unverified | report |
| 88 | InCoder 6.7B + CodeT | 34.4 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 89 | Llama 2 34B (0-shot) | 33 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 90 | Llama 2 13B (0-shot) | 30.6 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 91 | LLaMA 33B (0-shot) | 30.2 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 92 | InCoder 6.7B + MBR-Exec | 26.7 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 93 | InCoder 6.7B + Coder-Reviewer | 26.1 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 94 | InCoder 6.7B + Reviewer | 24.4 | – | Paper | Code | 2022 | 2 of 2 ran · 0 unverified | report |
| 95 | CodeGeeX-13B | 24.4 | – | Paper | Code | 2023 | 2 of 8 ran · 6 unverified | report |
| 96 | LLaMA 13B (0-shot) | 22 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 97 | Llama 2 7B (0-shot) | 20.8 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 98 | InCoder 6.7B (0-shot) | 19.4 | – | Paper | Code | 2022 | 6 of 10 ran · 4 unverified | report |
| 99 | LLaMA 7B (0-shot) | 17.7 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
All 99 rows shown. 98 link to a paper page on this site; 10 are marked as using additional training data in the archive. No GitHub stars are tracked; "Code" is the first repository the archive lists for the row. The archive carries no row tags, review links or community-submitted rows for this table; none are shown. archive 2025-07-28
Syntology Ran reads "N of M ran · U unverified": of the M code samples Syntology harvested from repositories linked to that row's paper (joined by arXiv id), N executed on a synthesized input and the other U = M−N are unverified (harvested, no recorded run). It counts code from repositories linked to that row's paper, not this result: the row's number was not reproduced and nothing here is a correctness claim. The other cell texts mean no graph line for the row: "linked, not harvested" (the archive links code, Syntology has not harvested it), "no code linked" (no code link in the archive), "not matched" (the row's paper URL matched no paper on this site). 75 rows have a graph line, from 22 distinct papers; 63 rows (18 papers) have at least one sample that ran. Counting each paper once: Syntology ran 158 of 285 samples; 127 unverified. Separately, 31 of those 285 are pointer-only (licence): the site points at that code rather than redistributing it, a licence property recorded for ran and unverified samples alike; each cell's tooltip carries the row's own pointer-only count. Read from the graph 2026-09-24. Per-sample status is on the paper page.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections