From 45624e9e8f2378e0094aa1eebf7af8b54221f660 Mon Sep 17 00:00:00 2001 From: steventohme Date: Wed, 20 May 2026 03:35:04 -0700 Subject: [PATCH 1/2] Update Weave Router scores to v0.62 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Weave Router v0.62 submission re-took #1 on the leaderboard. Updated the headline leaderboard entry with the new RouterArena eval results: - Arena Score: 74.61 → 76.09 - Accuracy: 78.43 → 79.32 - Cost per 1k: $0.92 → $0.61 - Robustness Score: 79.05 → 79.76 - Opt. Selection: 1.27 → 5.60 - Opt. Cost: 12.10 → 17.19 - Opt. Accuracy: 89.56 → 89.21 Co-Authored-By: Claude Opus 4.7 --- src/data/routerMetrics/leaderboard.json | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/src/data/routerMetrics/leaderboard.json b/src/data/routerMetrics/leaderboard.json index 0c1e958..c37adb6 100644 --- a/src/data/routerMetrics/leaderboard.json +++ b/src/data/routerMetrics/leaderboard.json @@ -12,14 +12,14 @@ }, { "Router Name": "Weave Router", - "Arena Score": 74.61, - "Optimal Selection Score": 1.27, - "Optimal Cost Score": 12.1, - "Optimal Acc. Score": 89.56, - "Robustness Score": 79.05, + "Arena Score": 76.09, + "Optimal Selection Score": 5.6, + "Optimal Cost Score": 17.19, + "Optimal Acc. Score": 89.21, + "Robustness Score": 79.76, "Latency Score": null, - "Accuracy": 78.43, - "Cost per 1k": 0.92 + "Accuracy": 79.32, + "Cost per 1k": 0.61 }, { "Router Name": "R2-Router", From 8749f3ac97b9b3302ff61432e82210fca2da26e4 Mon Sep 17 00:00:00 2001 From: yl231 Date: Wed, 20 May 2026 10:34:50 -0500 Subject: [PATCH 2/2] Update Weave Router metadata to v0.62 (modelPool + description) Mirrors the v0.62 submission config (RouterArena PR #102): - 12-model pool -> 7 models (drops claude-sonnet/haiku-4-5, gpt-5.4-mini, gpt-4.1, qwen3.5-flash-02-23, kimi-k2.5; adds qwen3-235b-a22b-2507) - description updated to mention v0.62 specifics: Jina v2 INT8 ONNX (768-dim) embeddings, top-p=4 cluster sum, k=160 clusters, alpha=0.25 Category scores and flip labels are intentionally left at v0.27-derived values for this PR; can be refreshed in a follow-up once v0.62 predictions are available locally. Co-Authored-By: Claude Opus 4.7 (1M context) --- src/data/routers.json | 9 ++------- src/data/routers.yaml | 9 ++------- 2 files changed, 4 insertions(+), 14 deletions(-) diff --git a/src/data/routers.json b/src/data/routers.json index 7e85b57..1a3498e 100644 --- a/src/data/routers.json +++ b/src/data/routers.json @@ -15,21 +15,16 @@ "Weave Router": { "name": "Weave Router", "type": "open-source", - "description": "Cluster-routing over a 12-model BYOK pool spanning Anthropic, OpenAI, Google, and OpenRouter providers. Embeds each prompt, scores against per-cluster model rankings, and selects the cost-quality optimum via an alpha-blended score.", + "description": "Cluster-routing (v0.62) over a 7-model BYOK pool spanning Anthropic, OpenAI, Google, and OpenRouter providers. Embeds each prompt with Jina v2 INT8 ONNX (768-dim), runs top-p=4 cluster sum against per-cluster rankings trained on the RouterArena full split with k=160 clusters, then selects the cost-quality optimum via an alpha-blended score (alpha=0.25).", "affiliation": "Weave", "modelPool": [ "claude-opus-4-7", - "claude-sonnet-4-5", - "claude-haiku-4-5", "gpt-5.5", - "gpt-5.4-mini", - "gpt-4.1", "gemini-3.1-pro-preview", "gemini-3.1-flash-lite-preview", "deepseek/deepseek-v4-pro", "deepseek/deepseek-v4-flash", - "qwen/qwen3.5-flash-02-23", - "moonshotai/kimi-k2.5" + "qwen/qwen3-235b-a22b-2507" ], "websiteUrl": "https://weaverouter.com", "githubUrl": "https://github.com/workweave/router" diff --git a/src/data/routers.yaml b/src/data/routers.yaml index dd7d89b..cee8bd0 100644 --- a/src/data/routers.yaml +++ b/src/data/routers.yaml @@ -1,21 +1,16 @@ Weave Router: name: Weave Router type: open-source - description: Cluster-routing over a 12-model BYOK pool spanning Anthropic, OpenAI, Google, and OpenRouter providers. Embeds each prompt, scores against per-cluster model rankings, and selects the cost-quality optimum via an alpha-blended score. + description: Cluster-routing (v0.62) over a 7-model BYOK pool spanning Anthropic, OpenAI, Google, and OpenRouter providers. Embeds each prompt with Jina v2 INT8 ONNX (768-dim), runs top-p=4 cluster sum against per-cluster rankings trained on the RouterArena full split with k=160 clusters, then selects the cost-quality optimum via an alpha-blended score (alpha=0.25). affiliation: Weave modelPool: - claude-opus-4-7 - - claude-sonnet-4-5 - - claude-haiku-4-5 - gpt-5.5 - - gpt-5.4-mini - - gpt-4.1 - gemini-3.1-pro-preview - gemini-3.1-flash-lite-preview - deepseek/deepseek-v4-pro - deepseek/deepseek-v4-flash - - qwen/qwen3.5-flash-02-23 - - moonshotai/kimi-k2.5 + - qwen/qwen3-235b-a22b-2507 websiteUrl: https://weaverouter.com githubUrl: https://github.com/workweave/router