Repository navigation
Default model 'claude-opus-5' is stale and scattered; setup offers no Codex backend despite README support #1936
Description
Target release
Unsure
How were you running Hyperloom?
Local
Optimization domain
Inference
Issue type
Environment / setup
Which phase failed?
Setup / baseline
Summary
- The default Claude model is hard-coded as
claude-opus-5, which is not a current model, in several places:src/hyperloom/common/llm_config.py:166(DEFAULT_CLAUDE_MODEL),agents/quantization/driver/runner.py(DEFAULT_MODEL),inference_optimizer/breakdown/reporters/llm_client.py(three literals), and the setup skill's default option. - The README lists "LLM backend: Claude, Codex" and the CLI has
--codex-model/OPENAI_BASE_URL/OPENAI_API_KEYsupport, buthyperloom-setuponly asks forANTHROPIC_BASE_URL,ANTHROPIC_API_KEYandCLAUDE_MODEL. A Codex-only user has no setup path.
Expected
- One owner for the default model id (the other call sites derive from it), updated to a current model.
- Setup first asks which backend (Claude or Codex), then collects that side's base URL, credential and model.
Activity
fix(llm): default to claude-opus-5-5 from one owner; keep claude-opus…
chaojhou commentedon Oct 17, 2026
Triage against main
7e57b1e8f.llm_config.DEFAULT_CLAUDE_MODEL(claude-opus-5-5).claude-opus-5remains a valid fallback. Leftover literals:install_kernel_tools.sh:182,skills/hyperloom-setup/SKILL.md:136,examples/README.md:80, andslurm/proxy.env.template(stillclaude-opus-5). A small separate PR should derive them or pin them with a test..envis authoritative and has noHYPERLOOM_LLM_MODE,inference_optimizer/assets/install_baremetal.shinfers anthropic-only from the Anthropic side alone (:2448-2452). It then unsets and deletes the OpenAI side whenever the hosts differ (:2491-2496,:2585-2593). So a user who picks Both, or who switches an existing Claude setup to Codex, getsOPENAI_BASE_URL/OPENAI_API_KEY/OPENAI_CUSTOM_HEADERSwiped (leavingCODEX_MODELorphaned) after the skill's read-back check has already passed.Needed for #1942: rebase onto main, dropping the already-merged #1940 commits and resolving the
SKILL.md:136conflict. Fix the inference in the backend: derive the mode from which sides of.envactually carry a URL and key, rather than adding newHYPERLOOM_LLM_MODEvalues that would have to be kept in sync with.env. Add a contract test: with.envholding both sides on different hosts, run the setup backend and assertOPENAI_BASE_URLandOPENAI_API_KEYsurvive alongsideCODEX_MODEL.Re-labelled
type:feature. Priority: P2. Assigning @jiejingzhangamd (author of #1942).github-actions commentedon Oct 19, 2026
This issue has been quiet for about 2 days since our last reply.
It will be closed automatically in 1 day if there is no further activity.
If you are still working on this, please leave a comment with any update — that will reset the timer.
github-actions commentedon Oct 20, 2026
This issue has been automatically closed because there was no response for about 3 days after our last comment.
If you would like to continue the discussion, please feel free to reopen this issue or leave a new comment — we will be happy to take another look.
docs(setup): let setup configure the Codex (OpenAI) backend