Skip to content

fix(defaults): replace gemini-3-flash-preview and gemini-2.5-flash with gemini-3.5-flash - #32372

Closed
hbentel wants to merge 1 commit into
NousResearch:mainfrom
hbentel:fix/gemini-model-defaults-35-flash
Closed

fix(defaults): replace gemini-3-flash-preview and gemini-2.5-flash with gemini-3.5-flash#32372
hbentel wants to merge 1 commit into
NousResearch:mainfrom
hbentel:fix/gemini-model-defaults-35-flash

Conversation

@hbentel

@hbentel hbentel commented May 26, 2026

Copy link
Copy Markdown
Contributor

Closes #32360.

Problem

Two stale model defaults shipped across the codebase:

The stable replacement for both is gemini-3.5-flash (GA, stable versioned ID).

Changes

Replaces every hardcoded occurrence in:

  • Provider plugins: gemini, openrouter, kilocode
  • agent/auxiliary_client.py, agent/gemini_native_adapter.py, agent/gemini_cloudcode_adapter.py
  • plugins/memory/hindsight/
  • trajectory_compressor.py
  • hermes_cli/config.py, setup.py, models.py, main.py, goals.py
  • tools/vision_tools.py, tools/web_tools.py
  • datagen-config-examples/trajectory_compression.yaml
  • Website docs (English + zh-Hans i18n)
  • website/static/api/model-catalog.json (regenerated via scripts/build_model_catalog.py)

Test-only uses of gemini-3-flash-preview that test model-specific routing behavior (e.g. the #25123 regression test that verifies the unversioned preview doesn't receive thinking_config) are intentionally preserved.

Test plan

  • scripts/run_tests.sh tests/agent/ tests/tools/ tests/hermes_cli/ tests/cli/ tests/run_agent/ — all pass (pre-existing failures in test_anthropic_adapter, test_gateway_wsl, test_gateway_service, test_service_manager are macOS environment issues unrelated to this change)
  • Updated test_provider_parity.py assertions to expect gemini-3.5-flash
  • Updated test_auxiliary_client.py assertions for pool and fallback defaults
  • Regenerated model-catalog.json; test_model_catalog.py passes

🤖 Generated with Claude Code

…faults with gemini-3.5-flash

Closes NousResearch#32360. gemini-3-flash-preview is an unversioned preview alias with
no stability guarantees, and gemini-2.5-flash reaches end-of-life October 16, 2026.
Updates all hardcoded defaults across providers, auxiliary client, memory plugins,
trajectory compressor, CLI, setup wizard, docs (en + zh-Hans), and the model
catalog manifest. Also fixes the model-catalog.json sync by regenerating it.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@alt-glitch alt-glitch added type/bug Something isn't working P2 Medium — degraded but workaround exists provider/gemini Google Gemini (AI Studio, Cloud Code) comp/cli CLI entry point, hermes_cli/, setup wizard comp/agent Core agent runtime: loop, agent_init, prompt builder, context-compression, responses endpoint comp/plugins Plugin system and bundled plugins labels May 26, 2026
@hbentel

hbentel commented Jun 14, 2026

Copy link
Copy Markdown
Contributor Author

Closing — this PR has been overtaken by main and its broad approach now conflicts with merged maintainer decisions.

Already landed on main via other PRs:

Conflicts with merged intent: The bulk of this diff replaces gemini-3-flash-previewgemini-3.5-flash across defaults, pickers, aux models, and docs. But #37606 (merged) deliberately restored gemini-3-flash-preview as a supported model, so removing it wholesale goes against that decision.

Residual value is marginal: what remains that's genuinely still on main and not counter to #37606 is a handful of gemini-2.5-flash fallback defaults (probe default, _create_chat_completion default, cloudcode default, hindsight summarization). gemini-2.5-flash isn't shut down until Oct 16 2026 and these are rarely-hit fallbacks. If that migration is still wanted closer to the deadline, it's better as a fresh, tightly-scoped PR (touching only those defaults, leaving gemini-3-flash-preview intact, and updating the coupled test_gemini_free_tier_setup_block assertion).

Happy to open that focused PR if/when it's useful.

@hbentel hbentel closed this Jun 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

comp/agent Core agent runtime: loop, agent_init, prompt builder, context-compression, responses endpoint comp/cli CLI entry point, hermes_cli/, setup wizard comp/plugins Plugin system and bundled plugins P2 Medium — degraded but workaround exists provider/gemini Google Gemini (AI Studio, Cloud Code) type/bug Something isn't working

Projects

None yet

Development

Successfully merging this pull request may close these issues.

fix: update default Gemini model from gemini-2.5-flash to gemini-3.5-flash (shutdown Oct 16 2026)

2 participants