# Overall MCP vs Hardcoded Comparison

Averaged across all projects: EmailResponder, GameBuilder, MarkdownValidator

## Framework-Level Comparison (All Models Averaged)

| Metric | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| exact_match | 0.50 | 0.51 | -0.00 |
| any_order_match | 0.54 | 0.54 | -0.00 |
| precision | 0.97 | 0.97 | -0.00 |
| recall | 0.91 | 0.91 | +0.00 |
| retry_rate | 0.03 | 0.04 | -0.02 |
| average_score | 0.60 | 0.60 | +0.00 |
| pass_rate | 0.00 | 0.00 | +0.00 |

---

## exact_match

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.66 | 0.63 | +0.03 |
| DeepSeek-V3-1 | 0.66 | 0.71 | -0.05 |
| GPT-4o-mini | 0.79 | 0.73 | +0.07 |
| GPT-5 | 0.67 | 0.67 | -0.00 |
| Gemini-2.5-flash | 0.07 | 0.11 | -0.04 |
| Gemini-2.5-flash-nothinking | 0.05 | 0.08 | -0.03 |
| Qwen3-235b | 0.63 | 0.63 | -0.00 |

## any_order_match

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.68 | 0.65 | +0.03 |
| DeepSeek-V3-1 | 0.74 | 0.75 | -0.01 |
| GPT-4o-mini | 0.87 | 0.78 | +0.09 |
| GPT-5 | 0.67 | 0.67 | -0.00 |
| Gemini-2.5-flash | 0.08 | 0.14 | -0.05 |
| Gemini-2.5-flash-nothinking | 0.06 | 0.11 | -0.04 |
| Qwen3-235b | 0.66 | 0.67 | -0.02 |

## precision

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.93 | 0.91 | +0.02 |
| DeepSeek-V3-1 | 0.98 | 0.99 | -0.01 |
| GPT-4o-mini | 0.97 | 0.99 | -0.02 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.95 | 0.94 | +0.01 |
| Gemini-2.5-flash-nothinking | 0.97 | 0.97 | +0.00 |
| Qwen3-235b | 0.99 | 0.99 | +0.00 |

## recall

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.96 | 0.96 | +0.01 |
| DeepSeek-V3-1 | 0.96 | 0.96 | -0.00 |
| GPT-4o-mini | 0.98 | 0.96 | +0.01 |
| GPT-5 | 0.94 | 0.94 | -0.00 |
| Gemini-2.5-flash | 0.79 | 0.78 | +0.01 |
| Gemini-2.5-flash-nothinking | 0.78 | 0.78 | -0.01 |
| Qwen3-235b | 0.94 | 0.94 | +0.00 |

## retry_rate

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.01 | 0.09 | -0.08 |
| DeepSeek-V3-1 | 0.01 | 0.01 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.10 | 0.11 | -0.01 |
| Gemini-2.5-flash-nothinking | 0.06 | 0.08 | -0.02 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

## average_score

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.79 | 0.82 | -0.03 |
| DeepSeek-V3-1 | 0.84 | 0.85 | -0.01 |
| GPT-4o-mini | 0.84 | 0.81 | +0.03 |
| GPT-5 | 0.70 | 0.72 | -0.03 |
| Gemini-2.5-flash | 0.06 | 0.11 | -0.05 |
| Gemini-2.5-flash-nothinking | 0.18 | 0.10 | +0.08 |
| Qwen3-235b | 0.83 | 0.78 | +0.04 |

## pass_rate

| Model | MCP (Avg) | Hardcoded (Avg) | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.00 | 0.00 | +0.00 |
| DeepSeek-V3-1 | 0.00 | 0.00 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 0.00 | 0.00 | +0.00 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

---

# EmailResponder: MCP vs Hardcoded

## Overall Summary (Averaged Across All Models)

| Metric | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| exact_match | 0.10 | 0.10 | -0.01 |
| any_order_match | 0.14 | 0.11 | +0.02 |
| precision | 0.96 | 0.95 | +0.01 |
| recall | 0.85 | 0.84 | +0.01 |
| retry_rate | 0.04 | 0.07 | -0.04 |
| pass_rate | 0.00 | 0.00 | +0.00 |

---

## exact_match

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.00 | 0.00 | +0.00 |
| DeepSeek-V3-1 | 0.06 | 0.26 | -0.20 |
| GPT-4o-mini | 0.56 | 0.30 | +0.26 |
| GPT-5 | 0.00 | 0.01 | -0.01 |
| Gemini-2.5-flash | 0.00 | 0.01 | -0.01 |
| Gemini-2.5-flash-nothinking | 0.02 | 0.07 | -0.05 |
| Qwen3-235b | 0.04 | 0.08 | -0.04 |

## any_order_match

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.03 | 0.02 | +0.01 |
| DeepSeek-V3-1 | 0.22 | 0.26 | -0.04 |
| GPT-4o-mini | 0.61 | 0.34 | +0.27 |
| GPT-5 | 0.00 | 0.01 | -0.01 |
| Gemini-2.5-flash | 0.00 | 0.01 | -0.01 |
| Gemini-2.5-flash-nothinking | 0.05 | 0.07 | -0.02 |
| Qwen3-235b | 0.05 | 0.09 | -0.04 |

## precision

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.80 | 0.74 | +0.06 |
| DeepSeek-V3-1 | 0.97 | 0.99 | -0.03 |
| GPT-4o-mini | 0.99 | 0.99 | +0.00 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.97 | 0.97 | +0.01 |
| Gemini-2.5-flash-nothinking | 0.96 | 0.96 | +0.00 |
| Qwen3-235b | 1.00 | 1.00 | -0.00 |

## recall

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.90 | 0.89 | +0.00 |
| DeepSeek-V3-1 | 0.88 | 0.88 | -0.00 |
| GPT-4o-mini | 0.94 | 0.89 | +0.04 |
| GPT-5 | 0.83 | 0.83 | -0.00 |
| Gemini-2.5-flash | 0.75 | 0.72 | +0.04 |
| Gemini-2.5-flash-nothinking | 0.80 | 0.82 | -0.01 |
| Qwen3-235b | 0.84 | 0.85 | -0.00 |

## retry_rate

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.04 | 0.28 | -0.24 |
| DeepSeek-V3-1 | 0.02 | 0.02 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.09 | 0.10 | -0.01 |
| Gemini-2.5-flash-nothinking | 0.12 | 0.12 | +0.00 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

## pass_rate

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.00 | 0.00 | +0.00 |
| DeepSeek-V3-1 | 0.00 | 0.00 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 0.00 | 0.00 | +0.00 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

# GameBuilder: MCP vs Hardcoded

## Overall Summary (Averaged Across All Models)

| Metric | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| exact_match | 0.66 | 0.65 | +0.02 |
| any_order_match | 0.71 | 0.71 | +0.00 |
| precision | 0.98 | 0.99 | -0.01 |
| recall | 0.94 | 0.93 | +0.00 |
| retry_rate | 0.00 | 0.01 | -0.01 |
| pass_rate | 0.00 | 0.00 | +0.00 |

---

## exact_match

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 1.00 | 0.88 | +0.12 |
| DeepSeek-V3-1 | 0.92 | 0.88 | +0.04 |
| GPT-4o-mini | 0.84 | 0.92 | -0.08 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 0.04 | 0.00 | +0.04 |
| Qwen3-235b | 0.84 | 0.84 | +0.00 |

## any_order_match

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 1.00 | 0.92 | +0.08 |
| DeepSeek-V3-1 | 1.00 | 1.00 | +0.00 |
| GPT-4o-mini | 1.00 | 1.00 | +0.00 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 0.04 | 0.08 | -0.04 |
| Qwen3-235b | 0.92 | 0.96 | -0.04 |

## precision

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 1.00 | 0.99 | +0.01 |
| DeepSeek-V3-1 | 0.96 | 0.98 | -0.02 |
| GPT-4o-mini | 0.92 | 0.99 | -0.06 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 1.00 | 0.97 | +0.03 |
| Qwen3-235b | 0.99 | 0.98 | +0.01 |

## recall

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 1.00 | 0.99 | +0.01 |
| DeepSeek-V3-1 | 1.00 | 1.00 | +0.00 |
| GPT-4o-mini | 1.00 | 1.00 | +0.00 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.76 | 0.74 | +0.02 |
| Gemini-2.5-flash-nothinking | 0.80 | 0.80 | +0.00 |
| Qwen3-235b | 0.99 | 0.99 | -0.01 |

## retry_rate

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.00 | 0.00 | +0.00 |
| DeepSeek-V3-1 | 0.00 | 0.00 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 0.00 | 0.08 | -0.08 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

## pass_rate

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.00 | 0.00 | +0.00 |
| DeepSeek-V3-1 | 0.00 | 0.00 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash-nothinking | 0.00 | 0.00 | +0.00 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

# MarkdownValidator: MCP vs Hardcoded

## Overall Summary (Averaged Across All Models)

| Metric | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| exact_match | 0.75 | 0.77 | -0.02 |
| any_order_match | 0.76 | 0.79 | -0.03 |
| precision | 0.98 | 0.98 | -0.00 |
| recall | 0.94 | 0.95 | -0.00 |
| retry_rate | 0.04 | 0.04 | +0.00 |
| average_score | 0.60 | 0.60 | +0.00 |

---

## exact_match

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.97 | 1.00 | -0.03 |
| DeepSeek-V3-1 | 1.00 | 1.00 | +0.00 |
| GPT-4o-mini | 0.98 | 0.97 | +0.02 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.20 | 0.32 | -0.12 |
| Gemini-2.5-flash-nothinking | 0.10 | 0.17 | -0.07 |
| Qwen3-235b | 1.00 | 0.97 | +0.03 |

## any_order_match

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 1.00 | 1.00 | +0.00 |
| DeepSeek-V3-1 | 1.00 | 1.00 | +0.00 |
| GPT-4o-mini | 1.00 | 1.00 | +0.00 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.25 | 0.40 | -0.15 |
| Gemini-2.5-flash-nothinking | 0.10 | 0.17 | -0.07 |
| Qwen3-235b | 1.00 | 0.97 | +0.03 |

## precision

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.99 | 1.00 | -0.01 |
| DeepSeek-V3-1 | 1.00 | 1.00 | +0.00 |
| GPT-4o-mini | 1.00 | 0.99 | +0.00 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.88 | 0.86 | +0.02 |
| Gemini-2.5-flash-nothinking | 0.95 | 0.97 | -0.02 |
| Qwen3-235b | 1.00 | 1.00 | +0.00 |

## recall

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 1.00 | 1.00 | +0.00 |
| DeepSeek-V3-1 | 1.00 | 1.00 | +0.00 |
| GPT-4o-mini | 1.00 | 1.00 | +0.00 |
| GPT-5 | 1.00 | 1.00 | +0.00 |
| Gemini-2.5-flash | 0.86 | 0.89 | -0.03 |
| Gemini-2.5-flash-nothinking | 0.73 | 0.74 | -0.01 |
| Qwen3-235b | 1.00 | 0.99 | +0.01 |

## retry_rate

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.00 | 0.00 | +0.00 |
| DeepSeek-V3-1 | 0.00 | 0.00 | +0.00 |
| GPT-4o-mini | 0.00 | 0.00 | +0.00 |
| GPT-5 | 0.00 | 0.00 | +0.00 |
| Gemini-2.5-flash | 0.20 | 0.22 | -0.02 |
| Gemini-2.5-flash-nothinking | 0.07 | 0.05 | +0.02 |
| Qwen3-235b | 0.00 | 0.00 | +0.00 |

## average_score

| Model | MCP | Hardcoded | Diff (MCP-Hardcoded) |
| --- | --- | --- | --- |
| DeepSeek-R1 | 0.79 | 0.82 | -0.03 |
| DeepSeek-V3-1 | 0.84 | 0.85 | -0.01 |
| GPT-4o-mini | 0.84 | 0.81 | +0.03 |
| GPT-5 | 0.70 | 0.72 | -0.03 |
| Gemini-2.5-flash | 0.06 | 0.11 | -0.05 |
| Gemini-2.5-flash-nothinking | 0.18 | 0.10 | +0.08 |
| Qwen3-235b | 0.83 | 0.78 | +0.04 |