From 314e573529897752c882be9aac4985931bcc26bc Mon Sep 17 00:00:00 2001 From: tin-berri Date: Tue, 8 Sep 2026 18:28:46 -0700 Subject: [PATCH] feat(auto-router): refresh family reasoning presets (#40341) --- .../public_endpoints/autorouter_presets.json | 43 +++++++++++++++---- .../src/lib/autorouter_presets.test.ts | 17 ++++---- 2 files changed, 42 insertions(+), 18 deletions(-) diff --git a/litellm/proxy/public_endpoints/autorouter_presets.json b/litellm/proxy/public_endpoints/autorouter_presets.json index 7d09db31127..7a251afc076 100644 --- a/litellm/proxy/public_endpoints/autorouter_presets.json +++ b/litellm/proxy/public_endpoints/autorouter_presets.json @@ -10,7 +10,12 @@ "REASONING": ["claude-opus-5"] }, "tier_model_configs": { - "REASONING": [{ "model_name": "claude-opus-5", "litellm_params": { "reasoning_effort": "high" } }] + "REASONING": [ + { + "model_name": "claude-opus-5", + "litellm_params": { "reasoning_effort": "high" } + } + ] }, "classifier_type": "heuristic_v2", "escalation_keywords": ["LITELLM ESCALATE"], @@ -23,16 +28,21 @@ }, "anthropic_family": { "label": "Anthropic Family", - "description": "Routes across the Claude model family: Haiku for simple queries, Sonnet for medium, Opus for complex, Opus at high thinking for reasoning.", + "description": "Routes across the Claude model family: Haiku for simple queries, Sonnet for medium, Opus for complex, Fable 5.1 at high thinking for reasoning.", "complexity_router_config": { "tiers": { "SIMPLE": ["claude-haiku-4-5"], "MEDIUM": ["claude-sonnet-5"], "COMPLEX": ["claude-opus-5"], - "REASONING": ["claude-opus-5"] + "REASONING": ["claude-fable-5-1"] }, "tier_model_configs": { - "REASONING": [{ "model_name": "claude-opus-5", "litellm_params": { "reasoning_effort": "high" } }] + "REASONING": [ + { + "model_name": "claude-fable-5-1", + "litellm_params": { "reasoning_effort": "high" } + } + ] }, "classifier_type": "heuristic", "escalation_keywords": ["LITELLM ESCALATE"], @@ -73,8 +83,18 @@ "REASONING": ["claude-opus-5"] }, "tier_model_configs": { - "MEDIUM": [{ "model_name": "muse-spark-1.2", "litellm_params": { "reasoning_effort": "xhigh" } }], - "COMPLEX": [{ "model_name": "kimi-k3", "litellm_params": { "reasoning_effort": "max" } }] + "MEDIUM": [ + { + "model_name": "muse-spark-1.2", + "litellm_params": { "reasoning_effort": "xhigh" } + } + ], + "COMPLEX": [ + { + "model_name": "kimi-k3", + "litellm_params": { "reasoning_effort": "max" } + } + ] }, "classifier_type": "llm", "classifier_llm_config": { @@ -93,16 +113,21 @@ }, "openai_family": { "label": "OpenAI Family", - "description": "Routes across the GPT model family: Luna for simple queries, Terra for medium, Sol for complex, Sol at xhigh thinking for reasoning.", + "description": "Routes across the GPT model family: Luna for simple queries, Terra for medium, Sol for complex, Astra at xhigh thinking for reasoning.", "complexity_router_config": { "tiers": { "SIMPLE": ["gpt-5.6-luna"], "MEDIUM": ["gpt-5.6-terra"], "COMPLEX": ["gpt-5.6-sol"], - "REASONING": ["gpt-5.6-sol"] + "REASONING": ["gpt-6-astra"] }, "tier_model_configs": { - "REASONING": [{ "model_name": "gpt-5.6-sol", "litellm_params": { "reasoning_effort": "xhigh" } }] + "REASONING": [ + { + "model_name": "gpt-6-astra", + "litellm_params": { "reasoning_effort": "xhigh" } + } + ] }, "classifier_type": "heuristic", "escalation_keywords": ["LITELLM ESCALATE"], diff --git a/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts b/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts index 5d1c7f73c71..8a5f83adbdf 100644 --- a/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts +++ b/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts @@ -149,13 +149,12 @@ describe("autorouter_presets", () => { ); }); - // Opus serves both tiers, so the effort is all that separates them and losing it fails silently. - it("pins the anthropic preset's reasoning tier to Opus at high thinking", () => { + it("pins the anthropic preset's reasoning tier to Fable 5.1 at high thinking", () => { const config = getPresetByKey("anthropic_family")!.complexity_router_config; expect(config.tiers.COMPLEX).toEqual(["claude-opus-5"]); - expect(config.tiers.REASONING).toEqual(["claude-opus-5"]); + expect(config.tiers.REASONING).toEqual(["claude-fable-5-1"]); expect(config.tier_model_configs).toEqual({ - REASONING: [{ model_name: "claude-opus-5", litellm_params: { reasoning_effort: "high" } }], + REASONING: [{ model_name: "claude-fable-5-1", litellm_params: { reasoning_effort: "high" } }], }); }); @@ -214,7 +213,7 @@ describe("autorouter_presets", () => { const preset = getPresetByKey("anthropic_family")!; const prefill = buildPresetPrefill(preset.complexity_router_config, groupsOnly(getRequiredModelsInPreset(preset))); expect(prefill.complexityRouterConfig.tier_model_params).toEqual({ - REASONING: { "claude-opus-5": { reasoning_effort: "high" } }, + REASONING: { "claude-fable-5-1": { reasoning_effort: "high" } }, }); }); @@ -227,21 +226,21 @@ describe("autorouter_presets", () => { }); }); - it("pins the OpenAI preset to the Luna, Terra, and Sol progression", () => { + it("pins the OpenAI preset to the Luna, Terra, Sol, and Astra progression", () => { const preset = getPresetByKey("openai_family")!; const expectedTiers = { SIMPLE: ["gpt-5.6-luna"], MEDIUM: ["gpt-5.6-terra"], COMPLEX: ["gpt-5.6-sol"], - REASONING: ["gpt-5.6-sol"], + REASONING: ["gpt-6-astra"], }; expect(preset.complexity_router_config.tiers).toEqual(expectedTiers); expect(preset.complexity_router_config.tier_model_configs).toEqual({ - REASONING: [{ model_name: "gpt-5.6-sol", litellm_params: { reasoning_effort: "xhigh" } }], + REASONING: [{ model_name: "gpt-6-astra", litellm_params: { reasoning_effort: "xhigh" } }], }); const prefill = buildPresetPrefill(preset.complexity_router_config, groupsOnly(getRequiredModelsInPreset(preset))); expect(prefill.complexityRouterConfig.tier_model_params).toEqual({ - REASONING: { "gpt-5.6-sol": { reasoning_effort: "xhigh" } }, + REASONING: { "gpt-6-astra": { reasoning_effort: "xhigh" } }, }); });