GPT-4o vs Claude 3.5 Haiku: Benchmark Comparison
Detailed comparison of GPT-4o and Claude 3.5 Haiku covering benchmarks, pricing, context window, and compliance.
Key Specifications
| Specification | GPT-4o | Claude 3.5 Haiku |
|---|---|---|
| Vendor | openai | anthropic |
| Version | 4o | 3.5-haiku |
| Release Date | 2024-05-13 | 2024-11-04 |
| Context Window | 128000 tokens | 200000 tokens |
| Input Modalities | text, image, audio | text, image |
| Output Modalities | text, audio | text |
| License | Proprietary | Proprietary |
| SOC2 | ✓ | ✓ |
| HIPAA | ✓ | ✗ |
| GDPR | ✓ | ✓ |
| ISO 27001 | ✓ | ✗ |
Benchmark Results
| Benchmark | GPT-4o | Claude 3.5 Haiku | Winner |
|---|---|---|---|
| ARC | — | 91.8 | Claude 3.5 Haiku |
| BBH | 83.1 | 70.2 | GPT-4o |
| GPQA | — | 31.1 | Claude 3.5 Haiku |
| GSM8K | 95.8 | 75.5 | GPT-4o |
| HUMANEVAL | 90.2 | 73.8 | GPT-4o |
| IFEVAL | — | 73.6 | Claude 3.5 Haiku |
| MATH | 76.6 | 53 | GPT-4o |
| MMLU | 88.7 | 77.6 | GPT-4o |
| MUSR | — | 52.3 | Claude 3.5 Haiku |
| WINOGRANDE | — | 83.8 | Claude 3.5 Haiku |
Pricing Comparison
| Tier (per Mtok) | GPT-4o | Claude 3.5 Haiku |
|---|---|---|
| Input | $2.5 | $0.8 |
| Output | $10 | $4 |
| Cache Read | $1.25 | $0 |
| Cache Write | $2.5 | $0 |
GPT-4o 対 Claude 3.5 Haiku
モデル概要
GPT-4o and Claude 3.5 Haiku are both notable options in the AI model market. This page compares their benchmarks, pricing, and compliance.
主要仕様
| ベンダー | リリース日 | コンテキストウィンドウ | ライセンス |
|---|---|---|---|
| Openai / Anthropic | 2024-05-13 / 2024-11-04 | 128K / 200K | Proprietary / Proprietary |
ベンチマークパフォーマンス
| ベンチマーク | GPT-4o | Claude 3.5 Haiku | 勝者 |
|---|---|---|---|
| ARC | — | 91.8 | B |
| BBH (BIG-Bench Hard) | 83.1 | 70.2 | A |
| GPQA | — | 31.1 | B |
| GSM8K (Grade School Math 8K) | 95.8 | 75.5 | A |
| HumanEval | 90.2 | 73.8 | A |
| IFEval | — | 73.6 | B |
| MATH | 76.6 | 53.0 | A |
| MMLU (Massive Multitask Language Understanding) | 88.7 | 77.6 | A |
| MUSR | — | 52.3 | B |
| WinoGrande | — | 83.8 | B |
料金比較
| 入力 | 出力 | キャッシュ読み取り | キャッシュ書き込み |
|---|---|---|---|
| — / — | — / — | — / — | — / — |
100万トークンあたり — A / B
強み & 弱み
GPT-4o
- ✅ MMLU score 88.7, strong knowledge reasoning.
- ✅ HumanEval 90.2, excellent code generation.
- ✅ GSM8K 95.8, robust math reasoning.
- ✅ 支持文本、图像、音频多模态输入。
- ⚠️ 闭源专有模型，不支持自托管。
Claude 3.5 Haiku
- ✅ 可靠的通用模型。
- ⚠️ 闭源专有模型，不支持自托管。
編集者コメント
GPT-4o and Claude 3.5 Haiku each have their strengths. Choose based on workload (code, long context, vision), referencing the tables above.
FAQ
Which model is better for coding tasks?
Refer to the HumanEval benchmark table; the model with a higher score is better suited for coding tasks.
Which model is cheaper?
Refer to the pricing comparison table above; the model with lower input/output prices is more cost-effective.
Which has a longer context window?
Refer to the key specifications table; the model with a larger context window is better for long documents.
参考文献
Editor's Take
See Editor's Take section.