diff --git a/litellm/proxy/public_endpoints/autorouter_presets.json b/litellm/proxy/public_endpoints/autorouter_presets.json index c2b13b81542..6a6642d4211 100644 --- a/litellm/proxy/public_endpoints/autorouter_presets.json +++ b/litellm/proxy/public_endpoints/autorouter_presets.json @@ -1,4 +1,26 @@ { + "1m_context": { + "label": "1M Context", + "description": "Routes across models with 1M-token context windows: Luna for simple queries, Terra for medium, Opus 5 for complex, Opus 5 at high thinking for reasoning.", + "complexity_router_config": { + "tiers": { + "SIMPLE": ["gpt-5.6-luna"], + "MEDIUM": ["gpt-5.6-terra"], + "COMPLEX": ["claude-opus-5"], + "REASONING": ["claude-opus-5"] + }, + "tier_model_configs": { + "REASONING": [{ "model_name": "claude-opus-5", "litellm_params": { "reasoning_effort": "high" } }] + }, + "classifier_type": "heuristic_v2", + "escalation_keywords": ["LITELLM ESCALATE"], + "classification_mode": "every_request", + "session_affinity": false, + "modality_routing": false, + "modality_pin_override": false, + "deployment_affinity": true + } + }, "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.", diff --git a/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py b/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py index 551e27a18f5..41439f28638 100644 --- a/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py +++ b/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py @@ -1104,6 +1104,16 @@ def test_get_autorouter_presets_local_mode_serves_bundled_catalog( assert response.status_code == 200 payload = response.json() assert "anthropic_family" in payload + assert payload["1m_context"]["complexity_router_config"]["classifier_type"] == "heuristic_v2" + assert payload["1m_context"]["complexity_router_config"]["tiers"] == { + "SIMPLE": ["gpt-5.6-luna"], + "MEDIUM": ["gpt-5.6-terra"], + "COMPLEX": ["claude-opus-5"], + "REASONING": ["claude-opus-5"], + } + assert payload["1m_context"]["complexity_router_config"]["tier_model_configs"] == { + "REASONING": [{"model_name": "claude-opus-5", "litellm_params": {"reasoning_effort": "high"}}] + } for preset in payload.values(): assert isinstance(preset["label"], str) assert isinstance(preset["description"], str) diff --git a/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx b/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx index 27a169524b9..836b4c6ff3d 100644 --- a/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx +++ b/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx @@ -636,7 +636,14 @@ describe("AddAutoRouterTab", () => { const labels = visibleOptions().map((option) => option.querySelector(".font-medium")?.textContent); - expect(labels).toEqual(["Anthropic Family", "Gemini Family", "Lite", "OpenAI Family", "Custom Configuration"]); + expect(labels).toEqual([ + "1M Context", + "Anthropic Family", + "Gemini Family", + "Lite", + "OpenAI Family", + "Custom Configuration", + ]); }); describe("routing test", () => { @@ -1060,7 +1067,14 @@ describe("AddAutoRouterTab", () => { expect(isOptionDisabled(optionByLabel("Anthropic Family")!)).toBe(false); }); const labels = visibleOptions().map((option) => option.querySelector(".font-medium")?.textContent); - expect(labels).toEqual(["Anthropic Family", "Gemini Family", "Lite", "OpenAI Family", "Custom Configuration"]); + expect(labels).toEqual([ + "Anthropic Family", + "1M Context", + "Gemini Family", + "Lite", + "OpenAI Family", + "Custom Configuration", + ]); }); it.each([ diff --git a/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts b/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts index 44ac6973a1c..a7333e4704b 100644 --- a/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts +++ b/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts @@ -28,7 +28,13 @@ const getPresetByKey = (key: string): AutoRouterPreset | undefined => PRESETS.fi describe("autorouter_presets", () => { it("hydrates exactly the bundled presets", () => { const presets = getAllPresets(); - expect(presets.map((p) => p.label).sort()).toEqual(["Anthropic Family", "Gemini Family", "Lite", "OpenAI Family"]); + expect(presets.map((p) => p.label).sort()).toEqual([ + "1M Context", + "Anthropic Family", + "Gemini Family", + "Lite", + "OpenAI Family", + ]); // Every preset carries all four fields the UI relies on; a JSON typo dropping one fails here. for (const p of presets) { expect(p).toMatchObject({ key: expect.any(String), label: expect.any(String), description: expect.any(String) }); @@ -202,6 +208,25 @@ describe("autorouter_presets", () => { }); }); + it("pins the 1M context preset to Luna, Terra, and Opus at high thinking", () => { + const preset = getPresetByKey("1m_context")!; + const expectedTiers = { + SIMPLE: ["gpt-5.6-luna"], + MEDIUM: ["gpt-5.6-terra"], + COMPLEX: ["claude-opus-5"], + REASONING: ["claude-opus-5"], + }; + expect(preset.complexity_router_config.classifier_type).toBe("heuristic_v2"); + expect(preset.complexity_router_config.tiers).toEqual(expectedTiers); + expect(preset.complexity_router_config.tier_model_configs).toEqual({ + REASONING: [{ model_name: "claude-opus-5", litellm_params: { reasoning_effort: "high" } }], + }); + const prefill = buildPresetPrefill(preset.complexity_router_config, groupsOnly(getRequiredModelsInPreset(preset))); + expect(prefill.complexityRouterConfig.tier_model_params).toEqual({ + REASONING: { "claude-opus-5": { reasoning_effort: "high" } }, + }); + }); + it("pins the gemini preset to concrete model ids, never Google's hot-swapping -latest aliases", () => { const gemini = getPresetByKey("gemini_family")!; const config = gemini.complexity_router_config;