ForgeCode

Models think. Agents act. Harnesses deliver outcomes. Proud to share that we’ve secured Rank #1, #2, and #4 on TermBench 2.0.

ForgeCode is now trending on GitHub in Rust And currently ranked #1 coding agent on TermBench.

ForgeCode is the #1 coding agent today. We have achieved 78.4% accuracy on TermBench 2.0, the global benchmark for terminal-based coding agents. We did this with a very small team. You could almost count us on a horse’s hoof! It took a lot of time, and at many points it felt like we were chasing a moving goalpost. These benchmarks are quite literally the Olympics of AI, where the world record resets every day. For now though, we’re the world champions here. And we’re not stopping. More on how we got the score in our blog (link below) This is just the beginning.

We are proud to announce the addition of new agents within the Forge ecosystem. We started with 𝐅𝐎𝐑𝐆𝐄, created to focus on implementation, and 𝐌𝐔𝐒𝐄, created to support analysis and planning. Building on this foundation, we are introducing three new agents that strengthen every stage of the development workflow:  𝐏𝐑𝐈𝐌𝐄 – Focused on documentation and blog review, ensuring clarity and accuracy  𝐏𝐀𝐑𝐊𝐄𝐑 – A technical writing expert who transforms complex ideas into engaging content  𝐒𝐀𝐆𝐄 – Specializing in codebase research and exploration to accelerate discovery With this broader lineup, Forgecode now supports builders across implementation, strategy, content, and research. The result is a more efficient, more complete environment for shipping projects.  Try our new agents now with npx forgecode@latest

We are excited to announce the launch of our Forge VSCode Extension VSCode Extension Features: • Integrate Forge AI into your workflow for smarter context handling and refactoring • Copy file references with line selection directly to your clipboard in Forge’s exact format:@[<filepath>:<line start>:<line end>] • Use the CTRL+U shortcut to instantly copy references with no manual formatting required • Automatic detection if Forge is not installed, with a guided setup process. This extension bridges your editor and Forge’s AI-powered CLI, helping you reference code accurately and move from idea to execution with fewer steps. Try Forge now at forgecode.dev

📣 Model update: OpenAI’s GPT-5 model family is now live on ForgeCode - GPT-5 - GPT-5 Mini - GPT-5 Nano What’s improved Early internal tests show stronger tool-calling reliability Steadier multi-step workflows for common dev tasks Reliability note In some cases, API endpoints intermittently return an empty response. We’re running a focused benchmark against Anthropic's Sonnet 4 and Opus 4.1 and will share results soon.

We’ve integrated OpenAI’s new GPT-OSS models, gpt-oss-20B and gpt-oss-120B, into Forgecode. Early impressions have been strong:  • Sub-second response times, even with complex, multi-file prompts  • High accuracy with CLI commands and tool usage  • Some quirks in collaboration, like stopping midway with “Here’s Phase 1” We’re refining prompts to improve multi-step follow-through. It’s exciting to have open-weight models that run locally and can be directly benchmarked against closed alternatives. This momentum fuels broader growth, encouraging other providers to release powerful open-source models and pushing the entire ecosystem forward. You can try the new models directly in your terminal at forgecode.dev We’d love to hear what you think

We’ve integrated OpenAI’s new GPT-OSS models, gpt-oss-20B and gpt-oss-120B, into Forgecode. Early impressions have been strong:  • Sub-second response times, even with complex, multi-file prompts  • High accuracy with CLI commands and tool usage  • Some quirks in collaboration, like stopping midway with “Here’s Phase 1” We’re refining prompts to improve multi-step follow-through. It’s exciting to have open-weight models that run locally and can be directly benchmarked against closed alternatives. This momentum fuels broader growth, encouraging other providers to release powerful open-source models and pushing the entire ecosystem forward. You can try the new models directly in your terminal at forgecode.dev We’d love to hear what you think