List your company

Downstream Task Accuracy via Paradigm Routing on Downstream Tasks (test)

73.4AccuracyOracle
22.9636.05549.1562.245Apr 8, 2026

Evaluation results

MethodLinks
OracleRouter Category=Oracle, Model=Gemini2026.0473.4—
OracleRouter Category=Oracle, Model=GPT-52026.0472.9—
OracleRouter Category=Oracle, Model=Qwen3-Max2026.0472.5—
Embedding (LR)Router Category=Emb, Model=GPT-52026.0464.2—
Combined (LR)Router Category=Emb, Model=GPT-52026.0464.2—
Best-singleRouter Category=Base, Model=GPT-52026.0462.4—
Handcrafted (MLP)Router Category=HC, Model=GPT-52026.0462.4—
Combined (MLP)Router Category=Emb, Model=GPT-52026.0462.4—
Combined (MLP)Router Category=Emb, Model=Gemini2026.0462.2—
Embedding (LR)Router Category=Emb, Model=Gemini2026.0461.9—
Combined (LR)Router Category=Emb, Model=Gemini2026.0461—
DirectRouter Category=Base, Model=GPT-52026.0460.3—
Self-routeRouter Category=ZS, Model=GPT-52026.0460.3—
Handcrafted (LR)Router Category=HC, Model=GPT-52026.0459.4—
Handcrafted (LR)Router Category=HC, Model=Gemini2026.0458.5—
Self-routeRouter Category=ZS, Model=Gemini2026.0456.8—
OracleRouter Category=Oracle, Model=Qwen3-30B2026.0456.8—
DirectRouter Category=Base, Model=Gemini2026.0455.5—
Best-singleRouter Category=Base, Model=Gemini2026.0455.5—
Handcrafted (MLP)Router Category=HC, Model=Gemini2026.0455.5—
Combined (LR)Router Category=Emb, Model=Qwen3-Max2026.0454.6—
Handcrafted (MLP)Router Category=HC, Model=Qwen3-Max2026.0453.3—
Embedding (LR)Router Category=Emb, Model=Qwen3-Max2026.0453.3—
Combined (MLP)Router Category=Emb, Model=Qwen3-Max2026.0453.3—
Handcrafted (LR)Router Category=HC, Model=Qwen3-Max2026.0451.5—
Best-singleRouter Category=Base, Model=Qwen3-Max2026.0450.7—
DirectRouter Category=Base, Model=Qwen3-Max2026.0449.8—
Self-routeRouter Category=ZS, Model=Qwen3-Max2026.0442.4—
Handcrafted (MLP)Router Category=HC, Model=Qwen3-30B2026.0437.1—
Handcrafted (LR)Router Category=HC, Model=Qwen3-30B2026.0434.9—
Combined (MLP)Router Category=Emb, Model=Qwen3-30B2026.0433.6—
Best-singleRouter Category=Base, Model=Qwen3-30B2026.0432.8—
Combined (LR)Router Category=Emb, Model=Qwen3-30B2026.0432.8—
Embedding (LR)Router Category=Emb, Model=Qwen3-30B2026.0432.3—
Self-routeRouter Category=ZS, Model=Qwen3-30B2026.0427.5—
DirectRouter Category=Base, Model=Qwen3-30B2026.0424.9—
Best-singleRouter Category=Base2026.04—50.3
Combined (LR)Router Category=Emb2026.04—53.1
Combined (MLP)Router Category=Emb2026.04—52.9
DirectRouter Category=Base2026.04—47.6
Embedding (LR)Router Category=Emb2026.04—52.9
Handcrafted (LR)Router Category=HC2026.04—51.1
Handcrafted (MLP)Router Category=HC2026.04—52.1
OracleRouter Category=Oracle2026.04—68.9
Self-routeRouter Category=ZS2026.04—46.7