v0.12.212 — Hide delisted/redirect aliases from /v1/models

Stop advertising delisted/redirect aliases in /v1/models (#187, thanks @KillerQueen-Z).

Model list no longer surfaces full-slug redirect aliases

MODEL_ALIASES mixes friendly short names (free, opus, gpt-120b) with full provider/model redirects for backward-compat + delisted models (e.g. free/deepseek-v4-pro → free/deepseek-v4-flash after V4 Pro's NVIDIA host hung 2026-04-30). ALIAS_MODELS turned every alias into a listable model, so dead slugs leaked into /v1/models (and any downstream picker) as if they were real, available models — a Codex picker built from /v1/models showed "DeepSeek V4 Pro (free)" that silently routes to v4-flash.

Fix: only surface friendly short aliases (no /). Full-slug redirect aliases are skipped from the advertised list — their target is a real model that's already listed, so nothing the user wants disappears. They remain fully callable via resolveModelAlias() (which reads MODEL_ALIASES directly), so pinned callers don't break. One-line filter in ALIAS_MODELS; MODEL_ALIASES and resolution logic are untouched.

• Picker + allowlist unaffected — every top-models.json entry is a real catalog id, never an alias key, so VISIBLE_OPENCLAW_MODELS / the OpenClaw /model picker / the openclaw.json allowlist sync see no change.
• 3 delisted redirects that are also real catalog ids (openai/o1-mini, nvidia/kimi-k2.5, google/gemini-3-pro-preview) drop off the HTTP discovery surface too — intended: all three are dead redirects whose live successors stay listed, and all stay callable.

Maintenance

• Formatted test/test-e2e.ts (a pre-existing prettier --check failure that was red-lining the Lint & Typecheck required check on every PR) and added .pytest_cache/ to .gitignore.

---

npm: @blockrun/clawrouter@0.12.212 · Full changelog: https://github.com/BlockRunAI/ClawRouter/blob/main/CHANGELOG.md

v0.12.211 — Z.AI GLM-5.2 flagship added

Add Z.AI's new flagship GLM-5.2 (launched 2026-06-16), aligning with BlockRun's source-of-truth catalog (zai/glm-5.2, first in the GLM lineup).

GLM-5.2 added

• zai/glm-5.2 added to the catalog — 1M-token context (262K max output), paid per-token at $1.40/$4.40 (same rate as GLM-5.1, cached $0.26). Z.AI's newest flagship: beats GPT-5.5 on long-horizon coding at a fraction of the cost. Categories: chat, reasoning, coding.
• Bare glm alias retargeted GLM-5.1 → GLM-5.2 (bare alias always tracks the newest flagship). New glm-5.2 explicit shorthand added. glm-5.1 explicit pin preserved as the 200K-context predecessor (identical price, so no cost-stability tradeoff — kept for behavioral parity).
• Picker visibility — added to top-models.json (head of the GLM group), so it surfaces in the OpenClaw /model picker and the allowlist sync on next provider refresh. README Mid-Range pricing table updated.
• No router-tier changes: GLM models are accessed by alias/direct call, not wired into any auto/eco/premium tier chain — GLM-5.2 follows the same path. All 619 tests pass.

---

npm: @blockrun/clawrouter@0.12.211

v0.12.210 — Free-tier realign with BlockRun 06-14 refresh

Align the free tier with BlockRun's 2026-06-14 free-tier refresh (blockrun commit 5817ecd: self-healing health gate + probe-verified NVIDIA lineup). BlockRun found only 7 of 17 "available" free models were actually being served and replaced the stale manual redirect list with a per-model circuit breaker.

Free models realigned with the probe-verified lineup

• free/qwen3-coder-480b retired — NVIDIA EOL'd it on 2026-06-14; BlockRun now redirects qwen3-coder → seed-oss-36b server-side. Dropped from the auto-pick cascade (FREE_MODELS), the picker (top-models.json), and the router COMPLEX backstops. The catalog entry + explicit nvidia/qwen3-coder-480b alias stay for pinned callers (BlockRun still resolves them via redirect); the generic coding shorthands (qwen-coder, glm-free, devstral, …) now point at the live successor free/seed-oss-36b.
• free/deepseek-v4-flash removed from the eco SIMPLE fallback — NVIDIA upstream hung; BlockRun redirects it. Replaced with the live free/qwen3-next-80b-a3b-instruct. Catalog entry kept (still direct-callable, redirects server-side).
• 6 new live free models added to the catalog + aliases (probe-verified by BlockRun): free/qwen3-next-80b-a3b-instruct (262K ctx, reasoning + coding), free/seed-oss-36b (coding), free/mistral-nemotron, free/step-3.7-flash (reasoning), free/nemotron-nano-9b-v2 (fast lightweight), free/nemotron-nano-12b-v2-vl (vision-language).
• Auto-pick cascade 7 → 12, gpt-oss kept at the head (heavy-user default — never retired). Picker free set 7 → 8 (qwen3-coder-480b → qwen3-next-80b-a3b-instruct + seed-oss-36b). README/SKILL free-count synced 7 → 8; install-script /model help echoes refreshed to live models.

Updater: clean stale OpenClaw install metadata (#186, thanks @0xCheetah1)

• scripts/update.sh now removes ClawRouter's stale record from ~/.openclaw/plugins/installs.json after a verified update. OpenClaw 2026.6 migrates its plugin install index toward shared SQLite; a leftover ClawRouter installRecords entry that no longer matches the installed version/path made openclaw doctor warn about conflicting install metadata on every run. The updater now compares the legacy record against the current package across the extensions/, project-scoped, and global npm layouts, and removes only ClawRouter's record (other plugins untouched) when it is stale — guarded by a full try/catch and an atomic write.

---

npm: @blockrun/clawrouter@0.12.210

v0.12.209 — Catalog sweep: Fable 5 revert, Kimi K2.7, free tier

Registry realignment with the 2026-06-13 → 06-14 BlockRun catalog sweep: Fable 5 delisting reverted, Kimi K2.7 promoted, free tier strengthened. 619 tests passing.

Fable 5 delisting (revert of the 06-11 promotion)

• anthropic/claude-fable-5 DELISTED by Anthropic 2026-06-13 (offer withdrawn upstream — no longer served on direct Anthropic or Bedrock). BlockRun removed the catalog entry and redirects fable → opus-4.8 (route.ts MODEL_REDIRECTS). v0.12.208's same-day promotion of Fable 5 to premium COMPLEX primary was therefore pointing default premium-tier complex traffic at a now-dead model. Reverted: catalog entry removed, all fable* aliases (fable, fable-5, fable-5.0, anthropic/fable, anthropic/claude-fable-5[.0]) now resolve to anthropic/claude-opus-4.8, and premium COMPLEX primary restored to opus-4.8 (its first fallback before the promotion). Pinned callers silently land on opus-4.8, matching the gateway.

Kimi K2.7 (BlockRun commit cd3d79b, 2026-06-13)

• moonshot/kimi-k2.7 added — Moonshot's new flagship: 256K context, multi-modal (image + video input), returns reasoning_content. AT-COST pricing $0.95/$4.00 — identical to K2.6 (BlockRun serves it via the OpenRouter credit pool failing over to direct Moonshot; zero margin by design). Bare aliases (kimi, moonshot, kimi-k2) retargeted K2.6 → K2.7; explicit kimi-k2.6 / kimi-k2.5 pins preserved. K2.6 marked superseded (hidden on BlockRun) but stays fully routable as an identical-cost first fallback.
• Router primaries promoted K2.6 → K2.7 (same price, so cost-neutral): auto.MEDIUM, agentic.MEDIUM, premium.SIMPLE. Each keeps K2.6 as its first fallback (in-family hot swap). K2.7 also prepended ahead of K2.6 in the premium MEDIUM/COMPLEX and agentic COMPLEX fallback chains. Picker (top-models.json) swapped K2.6 → K2.7.

Free tier strengthened (BlockRun 06-14 catalog sweep)

• Auto-pick set 6 → 7, gpt-oss kept at the head (heavy-user default, deliberately retained despite BlockRun hiding it). Mid/back strengthened with two BlockRun-re-featured free flagships: free/mistral-large-3-675b (675B general, un-retired — NVIDIA upstream recovered) and free/qwen3.5-122b-a10b (newest-gen Qwen). free/deepseek-v4-flash dropped from the auto-pick set + picker (BlockRun hid it) but stays catalog-routable for direct $0 calls. New free shorthands: mistral-large, qwen3.5-122b. eco SIMPLE fallback chain extended with both new models. README/SKILL free-count synced (6 → 7).

---

npm: @blockrun/clawrouter@0.12.209

v0.12.208 — Fix non-ASCII reasoning header crashing paid responses

Stop non-English prompts from crashing paid responses via the x-clawrouter-reasoning header.

• Non-ASCII routing reasoning crashed res.writeHead after settlement (src/proxy.ts). The routing reasoning string embeds matched keywords from the multilingual lists in src/router/config.ts (Cyrillic, CJK, …). Debug headers are on by default, so for e.g. Russian prompts the raw Cyrillic keywords landed in the x-clawrouter-reasoning response header and Node rejected the write with ERR_INVALID_CHAR — after the upstream call had completed and the x402 payment had settled. The client never received the body and retried, signing a fresh payment each round: a paid retry loop. Header values are now percent-encoded outside printable ASCII (sanitizeHeaderValue, reversible via decodeURIComponent), and as defense in depth a rejected writeHead sanitizes all header values and still delivers the paid body instead of throwing. (Test: proxy.reasoning-header.test.ts, 13 cases, including a live Russian-prompt repro through classifyByRules validated against Node's validateHeaderValue.)
• CLAWROUTER_DEBUG_HEADERS=off|false|0 — new env kill switch for the x-clawrouter-* debug headers, for clients that can't set the per-request x-clawrouter-debug: false header. Comments claiming the headers were "opt-in" corrected (they are on by default).

---

npm: @blockrun/clawrouter@0.12.208

v0.12.207 — Honor HTTPS_PROXY/HTTP_PROXY/ALL_PROXY for upstream

Honor standard proxy environment variables for upstream traffic.

• HTTPS_PROXY / HTTP_PROXY / ALL_PROXY fallback (src/upstream-proxy.ts). Node's fetch (undici) ignores the standard proxy env vars, so users whose system traffic is meant to flow through a local proxy (mihomo/clash without TUN mode, corporate proxies) had ClawRouter connecting directly while their curl tests went through the proxy — on throttled routes (e.g. RU → Google-hosted gateway) this surfaced as instant 500s on large request bodies, Premature close, and per-model timeouts, while small requests slipped through. When BLOCKRUN_UPSTREAM_PROXY is unset, ClawRouter now applies the standard env vars via undici's EnvHttpProxyAgent. NO_PROXY is honored, and loopback hosts (localhost, 127.0.0.1, ::1) are always excluded so local health checks and sibling proxies stay direct. BLOCKRUN_UPSTREAM_PROXY still wins when set; SOCKS URLs in the standard env vars are not auto-applied (a warning points to BLOCKRUN_UPSTREAM_PROXY=socks5://…). (Test: test/upstream-proxy.test.ts, 9 cases, dependency-injected — no process-global mutation in tests.)

---

npm: @blockrun/clawrouter@0.12.207

v0.12.159

• Market data tools — BlockRun gateway now exposes realtime and historical market data; ClawRouter wires them into OpenClaw as 6 first-class agent tools so the model stops scraping finance sites. Paid ($0.001 via x402, same wallet as LLM calls): blockrun_stock_price and blockrun_stock_history across 12 global equity markets (US, HK, JP, KR, UK, DE, FR, NL, IE, LU, CN, CA). Free (no x402 charge): blockrun_stock_list (ticker lookup / company-name search), blockrun_crypto_price (BTC-USD, ETH-USD, SOL-USD, …), blockrun_fx_price (EUR-USD, GBP-USD, JPY-USD, …), blockrun_commodity_price (XAU-USD gold, XAG-USD silver, XPT-USD platinum). Tool schemas advertise market codes, session hints (pre/post/on), and bar resolutions (1/5/15/60/240/D/W/M). Path routing extended: the partner-proxy whitelist in src/proxy.ts now matches /v1/(?:x|partner|pm|exa|modal|stocks|usstock|crypto|fx|commodity)/, routing all new paths through proxyPaidApiRequest (payFetch handles 402 when present, passes through 200 for free categories). Tool definitions added in src/partners/registry.ts; skills/clawrouter/SKILL.md gains a "Built-in Agent Tools" section listing market data + X intelligence + Polymarket alongside the LLM router.

v0.12.158

• SKILL.md data-flow + key-storage transparency — second-pass fix for the OpenClaw scanner on clawhub.ai. v0.12.157 cleared the original scanner flags (opaque credentials, implied multi-provider keys, no install artifact). A deeper rescan surfaced three new nuanced flags: prompts go to blockrun.ai (data-privacy framing), wallet private-key storage location/encryption undocumented, and users may expect strictly-local routing. All three addressed: description reframed as "Hosted-gateway LLM router" with explicit Ollama pointer for local-only needs; new Data Flow section with ASCII diagram, enumerated sent/not-sent lists, link to https://blockrun.ai/privacy; new Credentials & Local Key Storage section documenting config file paths per OS, 0600 POSIX permissions, plaintext parity with other OpenClaw provider keys, encryption guidance (FileVault/LUKS/BitLocker or burner wallet), src/wallet.ts source pointer; new Supply-Chain Integrity section with npm pack verification and tagged-release invariant.

v0.12.157

• SKILL.md credential transparency — rewrote skills/clawrouter/SKILL.md to clear the OpenClaw scanner's medium-confidence suspicious verdict on clawhub.ai. Frontmatter now declares repository: https://github.com/BlockRunAI/ClawRouter, license: MIT, and a structured metadata.openclaw.install array (kind: node, package: @blockrun/clawrouter, bins: [clawrouter]) so the registry entry has an auditable install artifact instead of a bare bash block. Body adds a Credentials & Data Handling section fully enumerating what models.providers.blockrun stores (walletKey / solanaKey — auto-generated locally, never transmitted; gateway / routing — non-sensitive), and explicitly states the plugin does not collect or forward third-party provider API keys (OpenAI/Anthropic/Google/DeepSeek/xAI/NVIDIA) — the blockrun.ai gateway owns those relationships and routes on the server side. Addresses the three scanner flags (opaque credential declaration, implied multi-provider credential collection, no install artifact for review) raised against v0.12.156 on https://clawhub.ai/1bcmax/clawrouter.

v0.12.156

v0.12.156 — Apr 20, 2026

• Kimi K2.6 added — Moonshot's new flagship (moonshot/kimi-k2.6, 256K context, vision + reasoning, $0.95 in / $4.00 out per 1M) registered in BLOCKRUN_MODELS with kimi-k2.6 alias. Added to the curated /model picker list (src/index.ts, scripts/update.sh, scripts/reinstall.sh), the README pricing table, docs/routing-profiles.md, and the AI-agent-facing model catalog in skills/clawrouter/SKILL.md. Premium routing tier (blockrun/premium) now uses K2.6 as the SIMPLE primary and as a fallback in MEDIUM/COMPLEX, with nvidia/kimi-k2.5 retained as the first fallback for reliability. The generic kimi/moonshot aliases still resolve to nvidia/kimi-k2.5 (matches BlockRun server's blockrun/kimi stance); users opt in to K2.6 explicitly via kimi-k2.6 or blockrun/premium.
• GitHub restored as canonical source — BlockRunAI GitHub org is back. package.json repository.url, README badges, CONTRIBUTING clone URL, openclaw.security.json, all docs (anthropic-*, clawrouter-cuts-*, clawrouter-vs-openrouter, 11-free-ai-models, llm-router-benchmark-*, smart-llm-router-14-dimension-classifier, subscription-failover, troubleshooting), skills/release/SKILL.md, and the sse-error-format regression-test comment now point at github.com/BlockRunAI/ClawRouter. GitLab mirror (gitlab.com/blockrunai/ClawRouter) is kept as a secondary remote for redundancy but is no longer advertised. Metadata + docs only; no runtime/code changes.

---