--- title: "Agents on Rails: Grok 4.6, GLM 5.3, Gemini 3.7 Flash, and Opus 4.8" url: https://daily.dev/posts/agents-on-rails-grok-4-6-glm-5-3-gemini-3-7-flash-and-opus-4-8-tvzhkpkxk source_url: https://rubyonrails.org/2026/8/17/agents-on-rails-grok-4-6-glm-5-3-gemini-3-7-flash-and-opus-4-8 type: article source: "Rails" published: 2026-08-17T19:36:36.224Z updated: 2026-08-17T19:37:25.823Z tags: ["ai", "ai-agents", "ruby", "rails", "claude"] reading_time: 4 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Agents on Rails: Grok 4.6, GLM 5.3, Gemini 3.7 Flash, and Opus 4.8 **[Rails](https://daily.dev/sources/rails)** · 4 min read · 0 upvotes · 0 comments ## Summary An updated benchmark report from the Rails team adds four new models—Grok 4.6, GLM 5.3, Gemini 3.7 Flash, and Claude Opus 4.8—to the Agents on Rails leaderboard, evaluating how coding agents perform on Rails tasks. Claude Opus 5 still leads with 58 of 63 runs solved, followed by Kimi K3 and Claude Fable 5. The team also analyzed model behavior traits (test-running frequency, verification steps, diff verbosity, comment placement) and found these habits don't predict solve rate. All 792 raw run traces from both rounds are now public on GitHub, along with a cautionary note: one model's full system recon exposed an OpenRouter API key in its trajectory logs, which was redacted before publishing. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Which AI model performs best on the Rails coding agent benchmark? Claude Opus 5 leads the Agents on Rails benchmark, solving 58 of 63 tasks, with Kimi K3 (open weights) and Claude Fable 5 trailing by one run. The benchmark ran 756 scored runs total, testing models like Grok 4.6, GLM 5.3, Gemini 3.7 Flash, and Claude Opus 4.8 on real Rails coding tasks at each provider's default effort level. _Compare AI coding agent performance on real framework tasks before picking one for daily.dev-tracked workflows._ ### Does writing more tests or running the test suite more often help an AI coding agent solve tasks correctly? No, these behavioral habits do not predict solve rate in Rails agent benchmarking. Each model works the same way regardless of task difficulty, and running the test suite 20 times performs no better than running it 3 times. Writing tests correlates with task difficulty rather than with actually winning the task. _Track findings like this on daily.dev when deciding how much to trust an agent's testing habits._ ### What security risk exists when running AI coding agents in a sandbox environment? Any secret present in an AI agent's environment can end up exposed in a recorded trajectory file. During a Rails benchmark, one model (Muse Spark 1.2) ran a full system recon command that leaked an OpenRouter API key into its run logs, requiring the team to redact key values across every file before publishing. _Follow agent-sandbox security lessons like this on daily.dev before exposing secrets to autonomous coding tools._ ## Similar posts on daily.dev - [Agents on Rails: the first benchmark report](https://daily.dev/posts/agents-on-rails-the-first-benchmark-report-1bsnzp6hz) · RUBYLAND · 2 upvotes · 2 comments - [More models, more choice: Comparing 11 different AI models](https://daily.dev/posts/more-models-more-choice-comparing-11-different-ai-models-uqrz8i7ag) · Hacker News · 1 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#ai-agents](https://daily.dev/tags/ai-agents), [#ruby](https://daily.dev/tags/ruby), [#rails](https://daily.dev/tags/rails), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/agents-on-rails-grok-4-6-glm-5-3-gemini-3-7-flash-and-opus-4-8-tvzhkpkxk)