Code Bison: Complete Benchmark Performance Guide

google code-bison 2023-05-10 Proprietary

In-depth analysis of Code Bison performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorgoogle
Versioncode-bison
Release Date2023-05-10
Context Window8192 tokens
Input Modalitiestext
Output Modalitiestext
LicenseProprietary
Documentationhttps://ai.google.dev/gemini-api/docs

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 65 % 2023-05-10 5-shot view
HUMANEVAL 76.2 pass@1 2023-05-10 — view
GSM8K 55 % 2023-05-10 0-shot CoT view
MATH 34.4 % 2023-05-10 0-shot CoT view
BBH 71.4 % 2023-05-10 3-shot CoT view
GPQA 29.1 % 2023-05-10 0-shot view
IFEVAL 59.2 % 2023-05-10 prompt_strict view
ARC 86.1 % 2023-05-10 challenge view
MUSR 49.8 % 2023-05-10 0-shot view
WINOGRANDE 77.9 % 2023-05-10 0-shot view

Pricing

Tier Price Currency
Input$0.5 / MtokUSD
Output$1 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://ai.google.dev/pricing · as of 2023-05-10

Compliance

  • Data Residency: US
  • SOC2: ✓
  • HIPAA: ✗
  • GDPR: ✓
  • ISO 27001: ✓

Code Bison

Model Overview

Google Vertex AI Code Bison 代码生成模型, 基于 PaLM 2, 8K 上下文, 支持多语言代码生成与补全。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Google code-bison 2023-05-10 8K text text Proprietary

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 65.0 % 5-shot
HumanEval 76.2 pass@1 —
GSM8K (Grade School Math 8K) 55.0 % 0-shot CoT
MATH 34.4 % 0-shot CoT
BBH (BIG-Bench Hard) 71.4 % 3-shot CoT
GPQA 29.1 % 0-shot
IFEval 59.2 % prompt_strict
ARC 86.1 % challenge
MUSR 49.8 % 0-shot
WinoGrande 77.9 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 8K is limited.

Use Cases

  • Code generation and debugging

References