Alma Outperforms Claude Code in Task Success Rate and Cost

iconChaincatcher
Share
AI summary iconSummary
According to Runta's FrontierHarness v1.1 benchmark, Alma outperforms Claude Code in task success rate and cost efficiency. Developed by Yetone, Alma achieved a 72.7% success rate with an average cost of $0.80 per task—16% cheaper than Claude Code. The test comprised 26,040 runs across 62 tasks. Alma supports local execution, code editing, and integrates with Herdr. As the crypto market remains volatile, developers are closely monitoring the Fear & Greed Index for signs of shifting sentiment.

ChainCatcher report: Alma, a local AI Agent developed independently by Yetone, ranked third in both task success rate and cost efficiency in the latest benchmark initiated by AI Agent infrastructure company Runta, outperforming tools such as Claude Code and OpenCode. The current snapshot shows Alma achieving a 72.7% task success rate, just behind Codex CLI at 77.9% and Pi at 72.8%, with Claude Code at 70.0%. Alma averages approximately $0.80 per successful task completion, compared to $0.94 for Claude Code—a 16% lower cost. This FrontierHarness v1.1 evaluation compares the performance of the same large models across different Agent execution frameworks, covering 21 Harness configurations, five large models, and 62 tasks, each repeated four times for a total of 26,040 runs—72 times the scale of the previous version. All tools were run in a unified cloud environment, measuring task success rate, token cost, and execution speed. Alma emphasizes local execution and long-term memory, enabling code reading and modification, terminal command execution, web searching, and support for Skills and MCP, with the ability to switch underlying models. Recently, Alma has integrated Herdr, allowing programming Agents like Claude Code, Codex, and Pi to continue running within Herdr while being centrally monitored through Alma’s unified chat interface for viewing execution progress, responding to queries, and approving actions. Users can also remotely manage multiple Agents on their computer via iPhone. The evaluation is still ongoing, and rankings may change; only open-source models such as GLM, DeepSeek, Kimi, and Muse are included, excluding proprietary models like GPT and Claude.

Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information. Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.