OpenHands v1.11.0 リリースOpenHands Releases v1.11.0

OpenHands, the open-source platform for autonomous software-development agents, has released version 1.11.0. Dated August 7, 2026, the update is a refinement-focused release whose most notable addition is per-run large language model cost tracking, a change that matters because agent-driven coding sessions frequently chain together many model calls whose combined token spend is hard to see until it accumulates.

The core new feature, contributed in pull request #16351, surfaces the LLM cost of each run directly in the Activity Log and carries that figure through to exports. In practical terms, a user can now look at an individual agent session and see roughly what it cost to run, rather than inferring spend from provider dashboards after the fact. Because OpenHands agents can call a model repeatedly as they read files, execute commands, and iterate on a task, per-run accounting gives a more granular view than an aggregate monthly total, and including it in exports appears aimed at teams that want to log or audit usage over time.

The release also reorders the Customize navigation (#16358) and polishes the user interface for Agent Canvas version updates (#16343). These are smaller, interface-focused changes rather than functional overhauls. The Agent Canvas is the visual workspace where users configure and interact with agents, and refining how version updates are presented is likely intended to make it clearer when a newer configuration or component is available. The published changelog also references additional home-screen work, though the available excerpt truncates the full list of entries.

Some background helps explain why cost tracking is a recurring theme in this category of tooling. OpenHands, formerly known as OpenDevin, is maintained by All Hands AI and belongs to a wave of software that lets an AI agent plan and carry out multi-step coding tasks: writing and editing code, running terminal commands, browsing documentation, and testing its own output inside a sandboxed environment. Unlike a single chat completion, an agent loop can issue dozens or hundreds of model requests to finish one task, so the economics differ substantially from conventional assistant use. That makes visibility into consumption a practical concern rather than a cosmetic one.

The move also reflects a broader industry trend toward LLM cost observability. As organizations adopt agents built on models from providers such as OpenAI, Anthropic, and Google, teams increasingly want to attribute spend to specific runs, users, or projects. Dedicated observability tools and libraries have emerged to trace token usage, latency, and cost across agent steps, and cloud providers expose their own billing dashboards. By embedding a per-run figure into the product itself, OpenHands reduces the need to correlate its activity with external billing data, at least for a first-order estimate. The accuracy of any in-app figure typically depends on up-to-date pricing for each model, so users running less common or self-hosted models may still want to verify numbers against their provider.

OpenHands sits alongside other agentic coding tools, including commercial offerings and IDE-integrated assistants, but its open-source nature means changes ship through public pull requests and are attributed to individual contributors, as the changelog entries above show. That development model tends to favor incremental, frequent releases, and version 1.11.0 fits that pattern: a set of targeted features rather than a single flagship capability.

For existing users, the update should be a low-friction upgrade, since the changes add information and adjust layout rather than alter core agent behavior. Those who care about budgeting or reporting are the most likely to benefit immediately from the cost figures, while the navigation and Agent Canvas tweaks are aimed at day-to-day usability. As always with release notes of this kind, the full set of changes, including bug fixes and internal adjustments not highlighted here, is documented in the project's changelog and associated pull requests for anyone who needs the complete picture.