JSON mode now exits 1 on failure, --session no longer eats your prompt in scripts, and the JSON stream is finally documented the way it actually behaves.
Elyra 0.9.54 adds a built-in oracle for second opinions, sessions that can find and read each other, commit trailers that lead back to the conversation, /handoff, and a safer approval rule for headless mode.
Gemini 4 Argon is now Elyra's Google default. Thinking levels now come from each model's own capabilities instead of name patterns, Vertex is generated from models.dev, and economy mode stops trusting the word batch.
Elyra 0.9.45 gates project configuration that can run code behind an explicit trust decision, and teaches Elyra to speak MCP in both directions.
Anthropic shipped Claude Opus 5.5, and Elyra now uses it as the default model. Why we switched, what it costs you, and how to keep Sonnet 5.
Five changes in Elyra 0.9.41 that stop the Anthropic prompt cache from being thrown away, add /bug and /system, and make session resume close to instant.
0.4.3 adds Jev, a small calibrated decision model that answers the agent's yes/no questions: which model tier a task needs, whether a shell command is destructive, and whether a finished run actually did what was asked.
Jev is a decision model: typed yes/no, choice and score questions that come back as probabilities for about two cents per thousand calls. Plus an opt-in bash gate, and the registry bug that quietly ate four models.
Elyra 0.9.38 shifts the model from having good tools to using them well: a working-method section in the system prompt, verification commands derived from your stack, smarter edit-failure handling, per-turn thinking levels, and pruning of stale tool output.
0.4.1 is about what the agent may not do: run a stranger's code, read your keys, upload your secrets. 0.4.2 is about what it's held to: a measured system prompt, and a turn that isn't done until the code compiles.