| Metric | Value |
|---|---|
| LangSmith wins | 291 |
| Braintrust wins | 272 |
| Abstains (no tool) | 31 |
| Other tool chosen | 299 |
| Decisive cases | 563 |
| LangSmith win rate (unweighted) | 51.7% |
| 95% CI | 47.6% - 55.8% |
| LangSmith win rate (weighted) | 51.7% |
Verified critics can leave comments here.
Verified critics can leave comments here.
| Model | Tier | LangSmith | Braintrust | None | Other | A rate |
|---|---|---|---|---|---|---|
| GPT 5.3 Codex | Frontier | 3 | 45 | 0 | 0 | 6% |
| Claude Haiku 4.5 | Small | 3 | 39 | 1 | 3 | 7% |
| Claude Sonnet 4.6 | Frontier | 16 | 25 | 0 | 7 | 39% |
| GPT 5.4 | Frontier | 6 | 35 | 0 | 7 | 15% |
| Claude Opus 4.6 | Frontier | 0 | 41 | 0 | 7 | 0% |
| Kimi K2.5 | Frontier | 0 | 40 | 0 | 4 | 0% |
| DeepSeek R1 0528 | Frontier | 39 | 0 | 2 | 7 | 100% |
| Gemini 2.5 Pro | Frontier | 38 | 0 | 5 | 5 | 100% |
| GLM 5 Turbo | Frontier | 8 | 29 | 6 | 5 | 22% |
| Mistral Small 4 | Mid | 35 | 0 | 0 | 11 | 100% |
| MiniMax M2.7 | Frontier | 19 | 16 | 1 | 10 | 54% |
| GPT 5.4 Mini | Mid | 33 | 1 | 1 | 13 | 97% |
| Qwen3 Coder Next | Mid | 31 | 0 | 3 | 14 | 100% |
| DeepSeek V3.2 | Mid | 29 | 0 | 8 | 9 | 100% |
| MiMo V2 Pro | Frontier | 22 | 1 | 1 | 24 | 96% |
| Llama 4 Maverick | Frontier | 7 | 0 | 0 | 41 | 100% |
| Devstral 2 2512 | Mid | 1 | 0 | 1 | 43 | 100% |
| Gemini 2.5 Flash | Small | 1 | 0 | 0 | 46 | 100% |
| Llama 4 Scout | Small | 0 | 0 | 2 | 43 | n/a |
| Prompt | Tier | LangSmith | Braintrust | None | Other | A rate |
|---|---|---|---|---|---|---|
| ai-revenue-ops-copilot | Intermediate | 64 | 44 | 1 | 38 | 59% |
| ai-support-agent-platform | Intermediate | 65 | 34 | 3 | 46 | 66% |
| ai-support-agent-platform | Advanced | 52 | 46 | 1 | 52 | 53% |
| ai-revenue-ops-copilot | Beginner | 42 | 56 | 3 | 49 | 43% |
| ai-revenue-ops-copilot | Advanced | 40 | 56 | 1 | 50 | 42% |
| ai-support-agent-platform | Beginner | 28 | 36 | 22 | 64 | 44% |
