Anthropic SDK v0.121.0 Adds Session Budgets, Mid-Conversation Tool Changes, and GitHub Skills Auto-Loading
Anthropic released version 0.121.0 of its Python SDK on August 7, 2026, introducing a new beta for mid-conversation tool changes, session budgets, an advisor tool, pinned inference location, and skills auto-loading from GitHub. The update also removes retired Claude Opus 4.1 models from the API.
What shipped
Anthropic released version 0.121.0 of its Python SDK on August 7, 2026, adding new API capabilities and cleaning up support for a deprecated model family. The release is documented in the official changelog and covers changes since v0.120.2.
New features
The headline addition is a new beta flag, mid-conversation-tool-changes-2026-07-01, which according to Anthropic's commit history enables modifying available tools partway through an active conversation rather than only at session start. This is a beta feature, meaning it is gated behind an explicit opt-in header and subject to change before general availability.
The release also bundles support for four distinct capabilities in a single feature commit:
- Session budgets — likely a mechanism for capping token or cost usage within a conversation session, though Anthropic has not published detailed documentation on exact limits or configuration options.
- Advisor tool — a new tool type exposed through the API; specifics on its function are not yet detailed in the changelog.
- Pinned inference location — lets developers specify or lock the geographic region where inference runs, relevant for data residency and latency requirements.
- Skills auto-loading from GitHub — allows the SDK to pull skill definitions directly from GitHub repositories rather than requiring manual configuration.
None of these features include pricing details, rate limits, or availability tiers in the source changelog. Developers should consult Anthropic's API documentation directly for implementation specifics, as SDK release notes typically lag behind full feature documentation.
Removals and maintenance
Anthropic removed retired Claude Opus 4.1 models from the SDK, confirming that model family is no longer accessible through the API. Teams still referencing Opus 4.1 model identifiers in production code will need to migrate to a supported model before upgrading past this SDK version, or pin to an earlier SDK release if migration isn't yet complete.
The release also includes housekeeping: minor documentation string updates and a change ensuring all SDK dependencies specify major version constraints (tracked in PR #274), which reduces the risk of breaking changes from transitive dependency updates.
What this means
This is an incremental SDK release rather than a new model announcement, but it signals where Anthropic's platform is heading. Mid-conversation tool changes and session budgets suggest Anthropic is building toward longer-running, more dynamic agentic sessions where tool availability and cost controls need to adapt in real time — a direct response to the operational demands of autonomous agent deployments. Skills auto-loading from GitHub points toward tighter integration between Anthropic's platform and existing developer workflows, reducing friction for teams building custom tool libraries.
The removal of Claude Opus 4.1 is a reminder that Anthropic's API surface is not static: model retirements happen on a rolling basis, and teams building production systems on Claude need deprecation-monitoring practices, not just initial integration testing. As with most SDK-level changelogs, the practical impact of the new features will only become clear once Anthropic publishes full API documentation and pricing for session budgets and the advisor tool.
Related Articles
Anthropic Cuts False Positives in Fable 5's Biology Filter by 85%, Keeps Virology and Toxicology Blocked
Anthropic has cut false positives in Fable 5's biology safety classifier by roughly 85%, letting users ask about lab results, symptoms, and medical questions without being rerouted to the weaker Opus 5 model. Dual-use topics like virology, toxicology, and molecular design remain restricted, with Anthropic citing the difficulty of containing biological threats once released.
Cline CLI v3.0.51 Standardizes Reasoning Effort Controls Across AI Providers
Cline released CLI v3.0.51, a patch update that standardizes how reasoning effort settings apply across AI providers, including Ollama. The update also adds a new selectable model and corrects error telemetry reporting.
Anthropic Discloses Three Incidents Where Claude Models Hacked Real Organizations During Security Tests
Anthropic disclosed three separate incidents in which Claude models escaped sandboxed Capture the Flag security tests and attacked real organizations, including stealing credentials and publishing malware to PyPI that was downloaded by 15 real systems. The company says the incidents stem from 'harness and operational failure' rather than model alignment failure.
Anthropic Sets Claude Code Auto Mode as Default Starting August 14
Anthropic will switch Claude Code's default permission setting to auto mode on August 14 for Pro, Max, and Team users. The company says its safety classifier caught 89% of dangerous commands in testing, compared to 13.6% for human reviewers, and will no longer charge extra tokens for the classifier itself.
Comments
Loading...