fix(providers): update stale Atlascloud/Ollama capability tests to the generic model-based path; add bare Moonshot/MiniMax/Z.ai model rows (kimi-k2.6, kimi-for-coding, minimax-m3, glm-5.1, glm-5v-turbo) mirroring vendor-prefixed rows; add Moonshot capability + engine capacity tests (#3023, #1310)

Co-Authored-By: Claude <noreply@anthropic.com> https://claude.ai/code/session_018zaP8vUfTAsrE38L6h6fw5
2026-06-11 00:17:53 +00:00
parent 139fe07d71
commit b9bcc9c710
3 changed files with 107 additions and 12 deletions
@@ -10756,14 +10756,30 @@ model = "deepseek-ai/deepseek-v4-pro"
    }

    #[test]
-    fn provider_capability_atlascloud_custom_model_is_chat_completions_without_thinking() {
+    fn provider_capability_atlascloud_v4_model_resolves_model_metadata() {
+        // #3023: Atlascloud uses the generic model-based path, so its default
+        // DeepSeek V4 model resolves the real V4 metadata instead of the old
+        // hardcoded legacy floor.
        let cap = provider_capability(ApiProvider::Atlascloud, "deepseek-ai/deepseek-v4-flash");
        assert_eq!(
            cap.context_window,
-            crate::models::LEGACY_DEEPSEEK_CONTEXT_WINDOW_TOKENS
+            crate::models::DEEPSEEK_V4_CONTEXT_WINDOW_TOKENS
        );
-        assert_eq!(cap.max_output, 4096);
-        assert!(!cap.thinking_supported);
+        assert_eq!(cap.max_output, 384_000);
+        assert!(cap.thinking_supported);
+        assert!(!cap.cache_telemetry_supported);
+        assert_eq!(
+            cap.request_payload_mode,
+            RequestPayloadMode::ChatCompletions
+        );
+    }
+
+    #[test]
+    fn provider_capability_moonshot_default_model_resolves_kimi_metadata() {
+        let cap = provider_capability(ApiProvider::Moonshot, DEFAULT_MOONSHOT_MODEL);
+        assert_eq!(cap.context_window, 262_144);
+        assert_eq!(cap.max_output, 262_144);
+        assert!(cap.thinking_supported);
        assert!(!cap.cache_telemetry_supported);
        assert_eq!(
            cap.request_payload_mode,
@@ -10788,8 +10804,26 @@ model = "deepseek-ai/deepseek-v4-pro"
    }

    #[test]
-    fn provider_capability_ollama_is_openai_compatible_without_thinking() {
+    fn provider_capability_ollama_deepseek_tag_uses_deepseek_heuristic() {
+        // #3023: known model families resolve through models.rs lookups even
+        // on Ollama — a legacy DeepSeek tag gets the 128K heuristic window.
        let cap = provider_capability(ApiProvider::Ollama, "deepseek-v3.1:671b");
+        assert_eq!(
+            cap.context_window,
+            crate::models::LEGACY_DEEPSEEK_CONTEXT_WINDOW_TOKENS
+        );
+        assert_eq!(cap.max_output, 4096);
+        assert!(!cap.thinking_supported);
+        assert!(!cap.cache_telemetry_supported);
+        assert_eq!(
+            cap.request_payload_mode,
+            RequestPayloadMode::ChatCompletions
+        );
+    }
+
+    #[test]
+    fn provider_capability_ollama_unknown_model_falls_back_to_8192() {
+        let cap = provider_capability(ApiProvider::Ollama, "llama3.2:3b");
        assert_eq!(cap.context_window, 8192);
        assert_eq!(cap.max_output, 4096);
        assert!(!cap.thinking_supported);
@@ -2863,6 +2863,33 @@ async fn pre_request_refresh_skips_compaction_below_normal_threshold() {
    assert_eq!(engine.session.messages.len(), before_len);
 }

+#[test]
+fn capacity_observation_uses_bare_kimi_context_window() {
+    // #3023: capacity math reads models::context_window_for_model directly,
+    // so bare Moonshot ids must resolve their real window, not the 128K
+    // legacy fallback.
+    let mut engine = build_engine_with_capacity(CapacityControllerConfig::default());
+    engine.session.model = "kimi-k2.6".to_string();
+    engine.session.messages.push(Message {
+        role: "user".to_string(),
+        content: vec![ContentBlock::Text {
+            text: "x".repeat(40_000),
+            cache_control: None,
+        }],
+    });
+
+    let estimated = engine.estimated_input_tokens() as f64;
+    let turn = TurnContext::new(1);
+    let observation = engine.capacity_observation(&turn);
+
+    let expected = estimated / 262_144.0;
+    assert!(
+        (observation.context_used_ratio - expected).abs() < 1e-9,
+        "context_used_ratio must use kimi-k2.6's 262,144-token window (got {})",
+        observation.context_used_ratio
+    );
+}
+
 #[tokio::test]
 async fn pre_request_refresh_invoked_when_medium_risk() {
    let capacity = CapacityControllerConfig {
@@ -263,9 +263,13 @@ fn known_context_window_for_model(model_lower: &str) -> Option<u32> {
        | "qwen/qwen3.6-27b"
        | "tencent/hy3-preview"
        | "moonshotai/kimi-k2.6"
-        | "moonshotai/kimi-k2.6:free" => Some(262_144),
-        "z-ai/glm-5.1" | "z-ai/glm-5v-turbo" => Some(202_752),
-        "minimax/minimax-m3" | "qwen/qwen3.6-flash" | "qwen/qwen3.6-plus" => Some(1_000_000),
+        | "moonshotai/kimi-k2.6:free"
+        | "kimi-k2.6"
+        | "kimi-for-coding" => Some(262_144),
+        "z-ai/glm-5.1" | "z-ai/glm-5v-turbo" | "glm-5.1" | "glm-5v-turbo" => Some(202_752),
+        "minimax/minimax-m3" | "minimax-m3" | "qwen/qwen3.6-flash" | "qwen/qwen3.6-plus" => {
+            Some(1_000_000)
+        }
        "xiaomi/mimo-v2.5-pro" | "xiaomi/mimo-v2.5" | "mimo-v2.5-pro" | "mimo-v2.5" => {
            Some(1_000_000)
        }
@@ -285,10 +289,12 @@ pub fn max_output_tokens_for_model(model: &str) -> Option<u32> {
        return Some(384_000);
    }
    match lower.as_str() {
-        "arcee-ai/trinity-large-thinking" | "trinity-large-thinking" | "moonshotai/kimi-k2.6" => {
-            Some(262_144)
-        }
-        "minimax/minimax-m3" => Some(524_288),
+        "arcee-ai/trinity-large-thinking"
+        | "trinity-large-thinking"
+        | "moonshotai/kimi-k2.6"
+        | "kimi-k2.6"
+        | "kimi-for-coding" => Some(262_144),
+        "minimax/minimax-m3" | "minimax-m3" => Some(524_288),
        "qwen/qwen3.6-35b-a3b" | "qwen/qwen3.6-27b" => Some(262_140),
        "qwen/qwen3.6-flash" | "qwen/qwen3.6-max-preview" | "qwen/qwen3.6-plus" => Some(65_536),
        "xiaomi/mimo-v2.5-pro" | "xiaomi/mimo-v2.5" | "mimo-v2.5-pro" | "mimo-v2.5" => {
@@ -322,7 +328,9 @@ pub fn model_supports_reasoning(model: &str) -> bool {
            | "google/gemma-4-26b-a4b-it:free"
            | "moonshotai/kimi-k2.6"
            | "moonshotai/kimi-k2.6:free"
+            | "kimi-k2.6"
            | "minimax/minimax-m3"
+            | "minimax-m3"
            | "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free"
            | "qwen/qwen3.6-flash"
            | "qwen/qwen3.6-35b-a3b"
@@ -335,6 +343,7 @@ pub fn model_supports_reasoning(model: &str) -> bool {
            | "mimo-v2.5-pro"
            | "mimo-v2.5"
            | "z-ai/glm-5.1"
+            | "glm-5.1"
    )
 }

@@ -587,6 +596,31 @@ mod tests {
        );
    }

+    #[test]
+    fn bare_provider_model_ids_mirror_vendor_prefixed_rows() {
+        // Direct-provider routes (Moonshot, MiniMax, Z.ai) serve bare model
+        // ids without the OpenRouter vendor prefix; both spellings must
+        // resolve identical metadata (#1310 ride-along on #3023).
+        for (model, expected_window) in [
+            ("kimi-k2.6", 262_144),
+            ("minimax-m3", 1_000_000),
+            ("glm-5.1", 202_752),
+        ] {
+            assert_eq!(context_window_for_model(model), Some(expected_window));
+            assert!(model_supports_reasoning(model));
+        }
+        assert_eq!(context_window_for_model("kimi-for-coding"), Some(262_144));
+        assert!(!model_supports_reasoning("kimi-for-coding"));
+        assert_eq!(context_window_for_model("glm-5v-turbo"), Some(202_752));
+        assert!(!model_supports_reasoning("glm-5v-turbo"));
+        assert_eq!(max_output_tokens_for_model("kimi-k2.6"), Some(262_144));
+        assert_eq!(
+            max_output_tokens_for_model("kimi-for-coding"),
+            Some(262_144)
+        );
+        assert_eq!(max_output_tokens_for_model("minimax-m3"), Some(524_288));
+    }
+
    #[test]
    fn deepseek_models_with_k_suffix_use_hint() {
        assert_eq!(context_window_for_model("deepseek-v3.2-32k"), Some(32_000));