Researchswe benchcode generationllmbenchmarks
SWE-bench Updates Bash-Only Coding Leaderboard With New Model Rankings
8.1
On 19 February 2026, SWE-bench published a fresh full run of its February 2025 'Bash Only' coding benchmark, evaluating models on 2,294 real-world problems drawn from 12 open-source repositories. Claude Opus 4.5 ranked first, followed by Gemini 3 Flash and MiniMax M2.5; OpenAI's GPT-5.2 placed sixth while GPT-5.3-Codex was absent, and the run used a uniform system prompt for fair comparison.
Key Points
- 1Ranks models: Claude Opus 4.5 first, Gemini 3 Flash second, MiniMax M2.5 third.
- 2Provides independent, non-self-reported benchmarking using identical system prompt across all evaluated models.
- 3Impacts model selection: highlights competitive Chinese models and notes GPT-5.3-Codex absence for API users.
Scoring Rationale
Independent, uniform benchmarking increases comparability and practical utility for model selection, but the Bash-only workload limits broader coding generalization.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
High-Value Overnight OrdersEasyDelivered International ShipmentsMediumOn-Time Delivery Rate by CarrierHard
250 free problems · No credit card
See all Logistics & Shipping problems
