Benchmarks · Coding · Agents & computer use
Ramp SWE-Bench v2 (Bash harness)
Ramp SWE-Bench v2's 500 production-derived tasks with every model in the same mini-swe-agent Bash harness, so the harness is held constant. Share resolved in a single attempt at high reasoning effort.
Category
22 models · 22 results · updated 2026-11-17 · labs.ramp.com ↗
Leaderboard
GPT-6.2 Astra high
Heavy line: best score at each price. Thin lines join one model's effort levels. Cost uses each model's most subsidized plan, or API price with no plan data. 7 more models under All models in settings.
More from this benchmark
2 published measuresTime per task22 results · lower is better
7 more models under All models in settings.
Output tokens per task22 results · lower is better
All results
22 rows| # | Model | Effort | Score | Subscription $ / task | API $ / task | Time | Output tokens | Harness |
|---|---|---|---|---|---|---|---|---|
| 1 | GPT-6.2 Astra | high | 75.0% ±3.8 | $0.068 | $1.03 | 8.2 min | 7.1k | mini-swe-agent |
| 2 | Gemini 4 Carbon | high | 74.0% ±3.8 | $0.32 | $5.33 | 1.1 h | 74k | mini-swe-agent |
| 3 | Fable 5.2 | high | 73.0% ±3.9 | $0.24 | $6.45 | 33 min | 63k | mini-swe-agent |
| 4 | GPT-6.2 Sol | high | 72.4% ±3.9 | $0.009 | $0.28 | 9.3 min | 8.1k | mini-swe-agent |
| 5 | Opus 5.5 | high | 71.2% ±4.0 | $0.013 | $0.99 | 14 min | 14k | mini-swe-agent |
| 6 | Gemini 4 Hydrogen | high | 70.8% ±4.0 | $0.071 | $1.75 | 28 min | 47k | mini-swe-agent |
| 7 | Fable 5.1 | high | 63.8% ±4.2 | $0.38 | $2.43 | 15 min | 18k | mini-swe-agent |
| 8 | GPT-6.1 Terra | high | 60.6% ±4.3 | $0.005 | $0.13 | 7.6 min | 7.7k | mini-swe-agent |
| 9 | Muse Wonder | high | 57.8% ±4.3 | $1.76 | $2.65 | 16 min | 26k | mini-swe-agent |
| 10 | Sonnet 5.5 | high | 57.6% ±4.3 | $0.003 | $0.24 | 6.8 min | 6.8k | mini-swe-agent |
| 11 | Grok 4.7 | high | 55.4% ±4.4 | $0.27 | $3.98 | 12 min | 18k | mini-swe-agent |
| 12 | Gemini 4 Argon | high | 54.8% ±4.4 | $4.34 | $6.51 | 1.1 h | 92k | mini-swe-agent |
| 13 | GPT-6 Astra | high | 54.6% ±4.4 | $0.060 | $1.69 | 9.5 min | 9.0k | mini-swe-agent |
| 14 | Muse Ember | high | 54.4% ±4.4 | $0.064 | $0.064 | 11 min | 29k | mini-swe-agent |
| 15 | DeepSeek V4.1 Flash 1017 | high | 53.4% ±4.4 | free | $0.16 | 16 min | 23k | mini-swe-agent |
| 16 | Haiku 5.5 | high | 52.8% ±4.4 | $0.002 | $0.12 | 18 min | 27k | mini-swe-agent |
| 17 | GPT-6.1 Sol | high | 52.0% ±4.4 | $0.009 | $0.27 | 8.8 min | 9.1k | mini-swe-agent |
| 18 | Gemini 3.8 Flash | high | 47.0% ±4.4 | $0.11 | $2.62 | 1.4 h | 81k | mini-swe-agent |
| 19 | Kimi K3.1 | high | 45.4% ±4.4 | $0.47 | $0.70 | 11 min | 12k | mini-swe-agent |
| 20 | GPT-6 Luna | high | 42.6% ±4.3 | $0.001 | $0.019 | 4.1 min | 11k | mini-swe-agent |
| 21 | Beam 1.0 | high | 34.0% ±4.1 | $0.081 | $0.081 | 7.8 min | 8.6k | mini-swe-agent |
| 22 | Nemotron 3.5 Super VL | high | 12.0% ±2.8 | free | $0.11 | 2.6 min | 12k | mini-swe-agent |