MBPP+

Expanded tests strengthen a collection of short Python programming problems.26 Oct 2026

One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
		Evidence	What the result shows
O1 Preview (Sept 2024) →OpenAI	80.2% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	Leads this displayed comparison. Source notes Source cell: 80.2
O1 Mini (Sept 2024) →OpenAI	78.8% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#2 of 89 in this displayed comparison; 1.4 pp below the leader. Source notes Source cell: 78.8
Qwen2.5-Coder-32B-Instruct →Qwen	77% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#3 of 89 in this displayed comparison; 3.2 pp below the leader. Source notes Source cell: 77.0
DeepSeek-Coder-V2-Instruct →DeepSeek	75.1% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#4 of 89 in this displayed comparison; 5.1 pp below the leader. Source notes Source cell: 75.1
Gemini 1.5 Pro 002 →Google	74.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#5 of 89 in this displayed comparison; 5.6 pp below the leader. Source notes Source cell: 74.6
Claude Sonnet 3.5 (June 2024) →Anthropic	74.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#6 of 89 in this displayed comparison; 5.9 pp below the leader. Source notes Source cell: 74.3
DeepSeek V2.5 (Nov 2024) →DeepSeek	74.1% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#7 of 89 in this displayed comparison; 6.1 pp below the leader. Source notes Source cell: 74.1
claude-3-opus (Mar 2024) →Anthropic	73.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#8 of 89 in this displayed comparison; 6.9 pp below the leader. Source notes Source cell: 73.3
GPT-4-Turbo (Nov 2023) →OpenAI	73.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#8 of 89 in this displayed comparison; 6.9 pp below the leader. Source notes Source cell: 73.3
DeepSeek V3 (Nov 2024) →DeepSeek	73% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#10 of 89 in this displayed comparison; 7.2 pp below the leader. Source notes Source cell: 73.0
GPT-4o (Aug 2024) →OpenAI	72.2% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#11 of 89 in this displayed comparison; 8 pp below the leader. Source notes Source cell: 72.2
GPT-4o Mini (July 2024) →OpenAI	72.2% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#11 of 89 in this displayed comparison; 8 pp below the leader. Source notes Source cell: 72.2
OpenCoder-8B-Instruct →	71.4% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#13 of 89 in this displayed comparison; 8.8 pp below the leader. Source notes Source cell: 71.4
DeepSeek-Coder-33B-instruct →DeepSeek	70.1% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#14 of 89 in this displayed comparison; 10.1 pp below the leader. Source notes Source cell: 70.1
GPT-3.5-Turbo (Nov 2023) →OpenAI	69.7% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#15 of 89 in this displayed comparison; 10.5 pp below the leader. Source notes Source cell: 69.7
Artigenz-Coder-DS-6.7B →	69.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#16 of 89 in this displayed comparison; 10.6 pp below the leader. Source notes Source cell: 69.6
claude-3-sonnet (Mar 2024) →Anthropic	69.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#17 of 89 in this displayed comparison; 10.9 pp below the leader. Source notes Source cell: 69.3
CodeQwen1.5-7B-Chat →Qwen	69% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#18 of 89 in this displayed comparison; 11.2 pp below the leader. Source notes Source cell: 69.0
Llama3-70B-instruct →	69% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#18 of 89 in this displayed comparison; 11.2 pp below the leader. Source notes Source cell: 69.0
Magicoder-S-DS-6.7B →	69% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#18 of 89 in this displayed comparison; 11.2 pp below the leader. Source notes Source cell: 69.0
claude-3-haiku (Mar 2024) →Anthropic	68.8% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#21 of 89 in this displayed comparison; 11.4 pp below the leader. Source notes Source cell: 68.8
OpenCodeInterpreter-DS-33B →	68.5% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#22 of 89 in this displayed comparison; 11.7 pp below the leader. Source notes Source cell: 68.5
Gemini 1.5 Flash 002 →Google	67.5% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#23 of 89 in this displayed comparison; 12.7 pp below the leader. Source notes Source cell: 67.5
WhiteRabbitNeo-33B-v1 →	66.9% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#24 of 89 in this displayed comparison; 13.3 pp below the leader. Source notes Source cell: 66.9
OpenCodeInterpreter-DS-6.7B →	66.4% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#25 of 89 in this displayed comparison; 13.8 pp below the leader. Source notes Source cell: 66.4
DeepSeek-Coder-6.7B-instruct →DeepSeek	65.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#26 of 89 in this displayed comparison; 14.6 pp below the leader. Source notes Source cell: 65.6
Grok Beta →xAI	65.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#26 of 89 in this displayed comparison; 14.6 pp below the leader. Source notes Source cell: 65.6
starcoder2-15b-instruct-v0.1 →	65.1% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#28 of 89 in this displayed comparison; 15.1 pp below the leader. Source notes Source cell: 65.1
XwinCoder-34B →	64.8% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#29 of 89 in this displayed comparison; 15.4 pp below the leader. Source notes Source cell: 64.8
code-millenials-34B →	64.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#30 of 89 in this displayed comparison; 15.6 pp below the leader. Source notes Source cell: 64.6
starchat2-15b-v0.1 →	64.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#30 of 89 in this displayed comparison; 15.6 pp below the leader. Source notes Source cell: 64.6
speechless-coder-ds-6.7B →	64.4% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#32 of 89 in this displayed comparison; 15.8 pp below the leader. Source notes Source cell: 64.4
Mixtral-8x22B-Instruct-v0.1 →Mistral	64.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#33 of 89 in this displayed comparison; 15.9 pp below the leader. Source notes Source cell: 64.3
Command-R+ →Cohere	63.5% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#34 of 89 in this displayed comparison; 16.7 pp below the leader. Source notes Source cell: 63.5
WaveCoder-Ultra-6.7B →	63.5% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#34 of 89 in this displayed comparison; 16.7 pp below the leader. Source notes Source cell: 63.5
WizardCoder-Python-34B-V1.0 →	63.2% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#36 of 89 in this displayed comparison; 17 pp below the leader. Source notes Source cell: 63.2
speechless-starcoder2-15b →	62.4% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#37 of 89 in this displayed comparison; 17.8 pp below the leader. Source notes Source cell: 62.4
DeepSeek-Coder-7B-instruct-v1.5 →DeepSeek	62.2% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#38 of 89 in this displayed comparison; 18 pp below the leader. Source notes Source cell: 62.2
Codestral-22B-v0.1 →Mistral	61.9% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#39 of 89 in this displayed comparison; 18.3 pp below the leader. Source notes Source cell: 61.9
Qwen1.5-72B-Chat →Qwen	61.6% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#40 of 89 in this displayed comparison; 18.6 pp below the leader. Source notes Source cell: 61.6
Gemini Pro 1.0 →Google	61.4% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#41 of 89 in this displayed comparison; 18.8 pp below the leader. Source notes Source cell: 61.4
speechless-codellama-34B-v2.0 →Meta	61.4% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#41 of 89 in this displayed comparison; 18.8 pp below the leader. Source notes Source cell: 61.4
CodeQwen1.5-7B →Qwen	60.8% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: False	#43 of 89 in this displayed comparison; 19.4 pp below the leader. Source notes Source cell: 60.8
Magicoder-S-CL-7B →	60.1% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#44 of 89 in this displayed comparison; 20.1 pp below the leader. Source notes Source cell: 60.1
Mistral Large (Mar 2024) →Mistral	59.5% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#45 of 89 in this displayed comparison; 20.7 pp below the leader. Source notes Source cell: 59.5
dolphin-2.6-mixtral-8x7b →Mistral	59% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#46 of 89 in this displayed comparison; 21.2 pp below the leader. Source notes Source cell: 59.0
DeepSeek-Coder-6.7B-base →DeepSeek	58.7% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: False	#47 of 89 in this displayed comparison; 21.5 pp below the leader. Source notes Source cell: 58.7
codegemma-7b-it →Google	56.9% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#48 of 89 in this displayed comparison; 23.3 pp below the leader. Source notes Source cell: 56.9
CodeLlama-34B →Meta	56.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: False	#49 of 89 in this displayed comparison; 23.9 pp below the leader. Source notes Source cell: 56.3
speechless-starcoder2-7b →	56.3% 1 report	MBPP+ · Pass@1[1]EvalPlus · official result data · Configuration: instruction prompt: True	#49 of 89 in this displayed comparison; 23.9 pp below the leader. Source notes Source cell: 56.3

Sources

Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation — published result table ↗
Liu, Jiawei, Xia, Chunqiu Steven, Wang, Yuyao, Zhang, Lingming · 2023
Dates & evaluation details
MBPP+ · Pass@1 · EvalPlus · official result data · 2026-10-26
Published model comparison, 2023. EvalPlus · official result data. Official EvalPlus leaderboard data, explicitly selected pass@1 field mbpp+. Expanded-test and original-test results remain separate. Missing scores are not inferred. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

What it measures

Expanded tests strengthen a collection of short Python programming problems.

How to read the result

Useful for checking basic implementation reliability and common edge cases under a repeatable execution harness.

Limits and comparison traps

Strong results on short functions do not establish long-horizon engineering capability; retain test-suite version and evaluation conditions.

Published evaluation

Official EvalPlus leaderboard data, explicitly selected pass@1 field mbpp+. Expanded-test and original-test results remain separate. Missing scores are not inferred.

Where this fits in an evaluation suite

The agent wrapper is part of the system. Pin tools, dependencies, timeouts, tests, and the repository revision. Compare contest coding separately from repository maintenance.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Official EvalPlus leaderboard data, explicitly selected pass@1 field mbpp+. Expanded-test and original-test results remain separate. Missing scores are not inferred.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation — published result table ↗
Liu, Jiawei, Xia, Chunqiu Steven, Wang, Yuyao, Zhang, Lingming · 2023
Primary benchmark documentation or model evaluation report ↗

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All code & engineering benchmarks →

Sources

What it measures

How to read the result

Limits and comparison traps

Published evaluation

Where this fits in an evaluation suite

A reproducible comparison

Continue exploring

AA-3D

Terminal-Bench 4.0

FrontierCode 1.1

Terminal-Bench 2.1