#12964: fix(#20156): Venice - add missing models and set a new model default

# What this PR fixes We've been getting reports from Venice users on OpenClaw where they can't see certain models, most critically Kimi K2.5 (kimi-k2-5). After investigating, the root cause is that the static model catalog (used as fallback when the Venice API is unreachable) was missing several models that are live on the API. Fixes #20156 ## Issues addressed - Kimi K2.5 invisible to users — kimi-k2-5 was not in the static catalog at all. When the /models API fetch fails (timeout, network issues), users fall back to the static list and Kimi K2.5 simply doesn't exist for them. - Missing models from catalog — Also added claude-opus-4-6 (Claude Opus 4.6) and zai-org-glm-4.7-flash (GLM 4.7 Flash), both live on the API but absent from the fallback. - Default model updated — Changed from llama-3.3-70b to kimi-k2-5 (strong reasoning + vision, fully private). - Wrong privacy classification — kimi-k2-thinking and minimax-m21 were listed as "anonymized" but the Venice API returns them as "private". Fixed to match reality. - Stale context window values — Updated all models to match the live API values. - Discovery reliability — Increased the API timeout from 5s → 8s, added filtering for offline models, and enriched catalog entries with live context window data when the API is reachable. Files changed (Venice-only, minimal scope) `src/agents/venice-models.ts` — catalog + discovery logic `src/commands/onboard-auth.config-core.ts` — default alias `docs/providers/venice.md` — model tables + recommendations `docs/providers/index.md` + models.md — default model refs Note: The CI test failure in `pi-embedded-helpers.isbillingerrormessage.test.ts` is a pre-existing issue on main introduced by #12946, unrelated to this PR. <!-- greptile_comment --> <h3>Greptile Summary</h3> Adds missing Venice models (`kimi-k2-5`, `kimi-k2-thinking`, `claude-opus-4-6`, `claude-sonnet-4-6`, `zai-org-glm-4.7-flash`, `zai-org-glm-5`, `minimax-m25`) to the static fallback catalog, changes the default model from `llama-3.3-70b` to `kimi-k2-5`, corrects privacy classifications for `kimi-k2-thinking` and `minimax-m21` (anonymized → private), and improves model discovery by filtering offline models, increasing the API timeout (5s → 8s), and enriching catalog entries with live context window data. - **Default model change**: `llama-3.3-70b` → `kimi-k2-5` consistently updated across code (`VENICE_DEFAULT_MODEL_ID`, onboard alias) and docs (`index.md`, `models.md`, `venice.md`). - **Discovery hardening**: Offline model filtering and `model_spec` null guard added. Previous review feedback about truthiness checks (`!= null` for `availableContextTokens`) and malformed entry handling (`m.model_spec &&` guard) have been addressed. - **Context window values**: Rounded down to match live Venice API values (e.g. 131072 → 128000, 262144 → 256000). - **Doc tables**: Fully rewritten to match the updated 31-model catalog (21 private + 10 anonymized), with correct context sizes and feature flags. <h3>Confidence Score: 4/5</h3> - This PR is safe to merge — it's a well-scoped data/config update with improved discovery robustness. - Changes are limited to a static model catalog, a default ID constant, and documentation. The discovery logic improvements (offline filtering, null guards, nullish coalescing) are correct and address prior review feedback. No behavioral regressions expected. Slight deduction because there are no tests for the Venice model discovery logic, so the new filtering/enrichment paths are only verified by inspection. - No files require special attention. All previous review comments have been addressed. _{Last reviewed commit: 2d0780b} <!-- greptile_other_comments_section --> <!-- /greptile_comment -->