Skip to content

Build and run agents you can see, understand and trust.

feat(models): add latest chat, TTS, and realtime model configs (#3488)

qbc6cc842350e82
diff --git a/src/agentscope/model/_anthropic/_models/claude-haiku-5-5.yaml b/src/agentscope/model/_anthropic/_models/claude-haiku-5-5.yaml
new file mode 100644
index 000000000..80ffdef46
--- /dev/null
+++ b/src/agentscope/model/_anthropic/_models/claude-haiku-5-5.yaml
@@ -0,0 +1,37 @@
+# Source: https://platform.claude.com/docs/en/models/haiku-5-5/overview
+name: claude-haiku-5-5
+label: Claude Haiku 5.5
+status: active
+
+input_types:
+ - text/plain
+ - application/x-thinking
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+ - application/x-thinking
+
+context_size: 1000000
+output_size: 128000
+
+parameter_overrides:
+ thinking_enable:
+ hidden: true
+ thinking_budget:
+ hidden: true
+ # Adaptive works at every effort level; older modes do not.
+ thinking_mode:
+ default: adaptive
+ anyOf:
+ - enum: [adaptive]
+ type: string
+ - type: "null"
+ thinking_display:
+ default: summarized
+ reasoning_effort:
+ default: medium
diff --git a/src/agentscope/model/_anthropic/_models/claude-opus-5-5.yaml b/src/agentscope/model/_anthropic/_models/claude-opus-5-5.yaml
new file mode 100644
index 000000000..b42cb6174
--- /dev/null
+++ b/src/agentscope/model/_anthropic/_models/claude-opus-5-5.yaml
@@ -0,0 +1,37 @@
+# Source: https://platform.claude.com/docs/en/models/opus-5-5/overview
+name: claude-opus-5-5
+label: Claude Opus 5.5
+status: active
+
+input_types:
+ - text/plain
+ - application/x-thinking
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+ - application/x-thinking
+
+context_size: 1000000
+output_size: 128000
+
+parameter_overrides:
+ thinking_enable:
+ hidden: true
+ thinking_budget:
+ hidden: true
+ # Adaptive works at every effort level; older modes do not.
+ thinking_mode:
+ default: adaptive
+ anyOf:
+ - enum: [adaptive]
+ type: string
+ - type: "null"
+ thinking_display:
+ default: summarized
+ reasoning_effort:
+ default: medium
diff --git a/src/agentscope/model/_anthropic/_models/claude-sonnet-5-5.yaml b/src/agentscope/model/_anthropic/_models/claude-sonnet-5-5.yaml
new file mode 100644
index 000000000..0adacc13e
--- /dev/null
+++ b/src/agentscope/model/_anthropic/_models/claude-sonnet-5-5.yaml
@@ -0,0 +1,37 @@
+# Source: https://platform.claude.com/docs/en/models/sonnet-5-5/overview
+name: claude-sonnet-5-5
+label: Claude Sonnet 5.5
+status: active
+
+input_types:
+ - text/plain
+ - application/x-thinking
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+ - application/x-thinking
+
+context_size: 1000000
+output_size: 128000
+
+parameter_overrides:
+ thinking_enable:
+ hidden: true
+ thinking_budget:
+ hidden: true
+ # Adaptive works at every effort level; older modes do not.
+ thinking_mode:
+ default: adaptive
+ anyOf:
+ - enum: [adaptive]
+ type: string
+ - type: "null"
+ thinking_display:
+ default: summarized
+ reasoning_effort:
+ default: high
diff --git a/src/agentscope/model/_openai_chat/_models/gpt-6-luna.yaml b/src/agentscope/model/_openai_chat/_models/gpt-6-luna.yaml
new file mode 100644
index 000000000..3da3ca1f3
--- /dev/null
+++ b/src/agentscope/model/_openai_chat/_models/gpt-6-luna.yaml
@@ -0,0 +1,29 @@
+# Source: https://developers.openai.com/api/docs/models/gpt-6-luna
+# Use Responses for agent tool calls; Chat has reasoning restrictions.
+name: gpt-6-luna
+label: GPT-6 Luna
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+
+context_size: 1050000
+output_size: 128000
+
+parameter_overrides:
+ # The adapter gates effort on thinking_enable, which is auto-hidden.
+ # Keep the provider's default reasoning until that gate is updated.
+ reasoning_effort:
+ hidden: true
+ temperature:
+ hidden: true
+ top_p:
+ hidden: true
diff --git a/src/agentscope/model/_openai_chat/_models/gpt-6-sol.yaml b/src/agentscope/model/_openai_chat/_models/gpt-6-sol.yaml
new file mode 100644
index 000000000..cbb517ebe
--- /dev/null
+++ b/src/agentscope/model/_openai_chat/_models/gpt-6-sol.yaml
@@ -0,0 +1,29 @@
+# Source: https://developers.openai.com/api/docs/models/gpt-6-sol
+# Use Responses for agent tool calls; Chat has reasoning restrictions.
+name: gpt-6-sol
+label: GPT-6 Sol
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+
+context_size: 1050000
+output_size: 128000
+
+parameter_overrides:
+ # The adapter gates effort on thinking_enable, which is auto-hidden.
+ # Keep the provider's default reasoning until that gate is updated.
+ reasoning_effort:
+ hidden: true
+ temperature:
+ hidden: true
+ top_p:
+ hidden: true
diff --git a/src/agentscope/model/_openai_chat/_models/gpt-6.1-sol.yaml b/src/agentscope/model/_openai_chat/_models/gpt-6.1-sol.yaml
new file mode 100644
index 000000000..5d23f4628
--- /dev/null
+++ b/src/agentscope/model/_openai_chat/_models/gpt-6.1-sol.yaml
@@ -0,0 +1,29 @@
+# Source: https://developers.openai.com/api/docs/models/gpt-6.1-sol
+# Use Responses for agent tool calls; Chat has reasoning restrictions.
+name: gpt-6.1-sol
+label: GPT-6.1 Sol
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+
+context_size: 1050000
+output_size: 128000
+
+parameter_overrides:
+ # The adapter gates effort on thinking_enable, which is auto-hidden.
+ # Keep the provider's default reasoning until that gate is updated.
+ reasoning_effort:
+ hidden: true
+ temperature:
+ hidden: true
+ top_p:
+ hidden: true
diff --git a/src/agentscope/model/_openai_response/_models/gpt-6-luna.yaml b/src/agentscope/model/_openai_response/_models/gpt-6-luna.yaml
new file mode 100644
index 000000000..4d85f2904
--- /dev/null
+++ b/src/agentscope/model/_openai_response/_models/gpt-6-luna.yaml
@@ -0,0 +1,26 @@
+# Source: https://developers.openai.com/api/docs/models/gpt-6-luna
+name: gpt-6-luna
+label: GPT-6 Luna
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+
+context_size: 1050000
+output_size: 128000
+
+parameter_overrides:
+ # The adapter gates effort on thinking_enable, which is auto-hidden.
+ # Keep the provider's default reasoning until that gate is updated.
+ reasoning_effort:
+ hidden: true
+ temperature:
+ hidden: true
diff --git a/src/agentscope/model/_openai_response/_models/gpt-6-sol.yaml b/src/agentscope/model/_openai_response/_models/gpt-6-sol.yaml
new file mode 100644
index 000000000..32fa00cfc
--- /dev/null
+++ b/src/agentscope/model/_openai_response/_models/gpt-6-sol.yaml
@@ -0,0 +1,26 @@
+# Source: https://developers.openai.com/api/docs/models/gpt-6-sol
+name: gpt-6-sol
+label: GPT-6 Sol
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+
+context_size: 1050000
+output_size: 128000
+
+parameter_overrides:
+ # The adapter gates effort on thinking_enable, which is auto-hidden.
+ # Keep the provider's default reasoning until that gate is updated.
+ reasoning_effort:
+ hidden: true
+ temperature:
+ hidden: true
diff --git a/src/agentscope/model/_openai_response/_models/gpt-6.1-sol.yaml b/src/agentscope/model/_openai_response/_models/gpt-6.1-sol.yaml
new file mode 100644
index 000000000..7bdc0ceda
--- /dev/null
+++ b/src/agentscope/model/_openai_response/_models/gpt-6.1-sol.yaml
@@ -0,0 +1,26 @@
+# Source: https://developers.openai.com/api/docs/models/gpt-6.1-sol
+name: gpt-6.1-sol
+label: GPT-6.1 Sol
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+ - image/gif
+ - image/webp
+ - application/pdf
+
+output_types:
+ - text/plain
+
+context_size: 1050000
+output_size: 128000
+
+parameter_overrides:
+ # The adapter gates effort on thinking_enable, which is auto-hidden.
+ # Keep the provider's default reasoning until that gate is updated.
+ reasoning_effort:
+ hidden: true
+ temperature:
+ hidden: true
diff --git a/src/agentscope/model/_volcengine/_models/doubao-seed-2-1-lite-260915.yaml b/src/agentscope/model/_volcengine/_models/doubao-seed-2-1-lite-260915.yaml
new file mode 100644
index 000000000..dcd451100
--- /dev/null
+++ b/src/agentscope/model/_volcengine/_models/doubao-seed-2-1-lite-260915.yaml
@@ -0,0 +1,19 @@
+# Source: https://www.volcengine.com/docs/82379/1330310
+name: doubao-seed-2-1-lite-260915
+label: Doubao Seed 2.1 Lite (260915)
+status: active
+
+# Only modalities implemented by the current Ark formatter are listed.
+input_types:
+ - text/plain
+ - application/x-thinking
+ - image/*
+ - video/*
+
+output_types:
+ - text/plain
+ - application/x-thinking
+
+# Use the same decimal k convention as the existing Ark cards.
+context_size: 1024000
+output_size: 256000
diff --git a/src/agentscope/model/_volcengine/_models/doubao-seed-evolving.yaml b/src/agentscope/model/_volcengine/_models/doubao-seed-evolving.yaml
new file mode 100644
index 000000000..8e717c860
--- /dev/null
+++ b/src/agentscope/model/_volcengine/_models/doubao-seed-evolving.yaml
@@ -0,0 +1,19 @@
+# Source: https://www.volcengine.com/docs/82379/1330310
+name: doubao-seed-evolving
+label: Doubao Seed Evolving
+status: active
+
+# Only modalities implemented by the current Ark formatter are listed.
+input_types:
+ - text/plain
+ - application/x-thinking
+ - image/*
+ - video/*
+
+output_types:
+ - text/plain
+ - application/x-thinking
+
+# Use the same decimal k convention as the existing Ark cards.
+context_size: 1024000
+output_size: 256000
diff --git a/src/agentscope/model/_xai/_models/grok-4.7.yaml b/src/agentscope/model/_xai/_models/grok-4.7.yaml
new file mode 100644
index 000000000..2a6b8ed6a
--- /dev/null
+++ b/src/agentscope/model/_xai/_models/grok-4.7.yaml
@@ -0,0 +1,18 @@
+# Source: https://docs.x.ai/developers/models/grok-4.7
+name: grok-4.7
+label: Grok 4.7
+status: active
+
+input_types:
+ - text/plain
+ - image/jpeg
+ - image/png
+
+output_types:
+ - text/plain
+ - application/x-thinking
+
+context_size: 500000
+# Local conservative cap, consistent with existing Grok cards.
+# The official model page does not publish a maximum output limit.
+output_size: 32000
diff --git a/src/agentscope/realtime/_gemini/_models/gemini-3.8-live.yaml b/src/agentscope/realtime/_gemini/_models/gemini-3.8-live.yaml
new file mode 100644
index 000000000..892f9927a
--- /dev/null
+++ b/src/agentscope/realtime/_gemini/_models/gemini-3.8-live.yaml
@@ -0,0 +1,31 @@
+# Source: https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live
+type: realtime_model
+name: gemini-3.8-live
+label: Gemini 3.8 Live
+status: active
+
+# Image input is not implemented by the current realtime adapter.
+input_types:
+ - audio/pcm
+ - text/plain
+output_types:
+ - audio/pcm
+ - text/plain
+
+input_sample_rate: 16000
+output_sample_rate: 24000
+
+supports_tools: true
+max_context_tokens: 131072
+# Audio-only duration without context-window compression.
+max_session_duration_s: 900
+
+# Source: https://ai.google.dev/gemini-api/docs/live-api/capabilities
+parameter_overrides:
+ voice:
+ enum: ["Zephyr", "Puck", "Charon", "Kore", "Fenrir", "Leda", "Orus",
+ "Aoede", "Callirrhoe", "Autonoe", "Enceladus", "Iapetus",
+ "Umbriel", "Algieba", "Despina", "Erinome", "Algenib",
+ "Rasalgethi", "Laomedeia", "Achernar", "Alnilam", "Schedar",
+ "Gacrux", "Pulcherrima", "Achird", "Zubenelgenubi",
+ "Vindemiatrix", "Sadachbia", "Sadaltager", "Sulafat"]
diff --git a/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.0-tts-flash.yaml b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.0-tts-flash.yaml
new file mode 100644
index 000000000..2171971d0
--- /dev/null
+++ b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.0-tts-flash.yaml
@@ -0,0 +1,27 @@
+# Source: https://help.aliyun.com/zh/model-studio/tts-model
+# Uses tts_v2.SpeechSynthesizer, not the Qwen3-TTS protocol.
+name: qwen-audio-3.0-tts-flash
+label: Qwen Audio 3.0 TTS Flash
+status: active
+
+input_types:
+ - text/plain
+
+output_types:
+ - audio/wav
+
+# Source: https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list
+# Curated system voices; explicitly pass voice when constructing the model.
+voices:
+ - longanfengyue
+ - longanyuanfei
+ - longanlingxi
+ - longanxiaoxin
+ - longanhuan_v3.6
+ - longjielidou_v3.6
+ - longpaopao_v3.6
+ - longhuohuo_v3.6
+ - longchuanshu_v3.6
+ - loongmary
+ - loongeva_v3.6
+ - loongjohn
diff --git a/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.0-tts-plus.yaml b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.0-tts-plus.yaml
new file mode 100644
index 000000000..4ef677ab3
--- /dev/null
+++ b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.0-tts-plus.yaml
@@ -0,0 +1,17 @@
+# Source: https://help.aliyun.com/zh/model-studio/tts-model
+# Uses tts_v2.SpeechSynthesizer, not the Qwen3-TTS protocol.
+name: qwen-audio-3.0-tts-plus
+label: Qwen Audio 3.0 TTS Plus
+status: active
+
+input_types:
+ - text/plain
+
+output_types:
+ - audio/wav
+
+# Source: https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list
+# Curated system voices; explicitly pass voice when constructing the model.
+voices:
+ - longanlingxin
+ - longanlufeng
diff --git a/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.1-tts-flash.yaml b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.1-tts-flash.yaml
new file mode 100644
index 000000000..9d3e8a48f
--- /dev/null
+++ b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.1-tts-flash.yaml
@@ -0,0 +1,19 @@
+# Source: https://help.aliyun.com/zh/model-studio/tts-model
+# Uses tts_v2.SpeechSynthesizer, not the Qwen3-TTS protocol.
+name: qwen-audio-3.1-tts-flash
+label: Qwen Audio 3.1 TTS Flash
+status: active
+
+input_types:
+ - text/plain
+
+output_types:
+ - audio/wav
+
+# Source: https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list
+# Curated system voices; explicitly pass voice when constructing the model.
+voices:
+ - longanhuan_v3.1
+ - longanlingxin_v3.1
+ - longanfengyue_v3.1
+ - xunanchuan_v3.1
diff --git a/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.1-tts-plus.yaml b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.1-tts-plus.yaml
new file mode 100644
index 000000000..fb098fe7d
--- /dev/null
+++ b/src/agentscope/tts/_dashscope/_cosyvoice_models/qwen-audio-3.1-tts-plus.yaml
@@ -0,0 +1,19 @@
+# Source: https://help.aliyun.com/zh/model-studio/tts-model
+# Uses tts_v2.SpeechSynthesizer, not the Qwen3-TTS protocol.
+name: qwen-audio-3.1-tts-plus
+label: Qwen Audio 3.1 TTS Plus
+status: active
+
+input_types:
+ - text/plain
+
+output_types:
+ - audio/wav
+
+# Source: https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list
+# Curated system voices; explicitly pass voice when constructing the model.
+voices:
+ - longanhuan_v3.1
+ - longanlingxin_v3.1
+ - longanfengyue_v3.1
+ - xunanchuan_v3.1
diff --git a/tests/realtime_gemini_test.py b/tests/realtime_gemini_test.py
index 85a4084c1..248443371 100644
--- a/tests/realtime_gemini_test.py
+++ b/tests/realtime_gemini_test.py
@@ -41,54 +41,17 @@ def _dump(events: list) -> list:
class GeminiCardsTest(unittest.TestCase):
"""The shipped cards and the credential lookup that finds them."""
- def test_cards(self) -> None:
- """Both Live dialog models, tagged with the adapter type."""
- self.assertListEqual(
- [
- (
- c.name,
- c.model_type,
- c.supports_tools,
- c.max_context_tokens,
- c.max_session_duration_s,
- c.output_sample_rate,
- )
- for c in GeminiRealtimeModel.list_models()
- ],
- [
- (
- "gemini-2.5-flash-native-audio-preview-12-2025",
- "gemini_realtime",
- True,
- 131072,
- 900,
- 24000,
- ),
- (
- "gemini-3.1-flash-live-preview",
- "gemini_realtime",
- True,
- 131072,
- 900,
- 24000,
- ),
- ],
- )
-
def test_credential_maps_card_back_to_class(self) -> None:
"""The service-layer lookup: card.model_type -> class, no scan."""
+ cards = GeminiRealtimeModel.list_models()
+ self.assertTrue(cards)
classes = {c.type: c for c in CRED.get_realtime_model_classes()}
self.assertDictEqual(
{
- card.name: classes[card.model_type].__name__
+ card.name: classes[card.model_type]
for card in CRED.list_realtime_models()
},
- {
- "gemini-2.5-flash-native-audio-preview-12-2025": (
- "GeminiRealtimeModel"
- ),
- "gemini-3.1-flash-live-preview": "GeminiRealtimeModel",
- },
+ {card.name: GeminiRealtimeModel for card in cards},
)
def test_unknown_model_name_is_rejected(self) -> None:
@@ -118,6 +81,17 @@ class GeminiCardsTest(unittest.TestCase):
class GeminiSetupTest(unittest.TestCase):
"""The setup message sent on connect."""
+ def test_gemini_38_setup(self) -> None:
+ """The new Live model reuses the supported audio session setup."""
+ expected = GeminiRealtimeModel(MODEL, CRED)._setup(
+ "be nice",
+ TOOLS,
+ "",
+ )
+ expected["setup"]["model"] = "models/gemini-3.8-live"
+ model = GeminiRealtimeModel("gemini-3.8-live", CRED)
+ self.assertDictEqual(model._setup("be nice", TOOLS, ""), expected)
+
def test_automatic_turn_detection(self) -> None:
"""Server VAD, tools and both transcriptions."""
model = GeminiRealtimeModel(MODEL, CRED)
diff --git a/tests/tts_dashscope_test.py b/tests/tts_dashscope_test.py
index 729301f4f..4c59b20f1 100644
--- a/tests/tts_dashscope_test.py
+++ b/tests/tts_dashscope_test.py
@@ -597,48 +597,29 @@ class TestDashScopeCosyVoiceTTSModel(IsolatedAsyncioTestCase):
synth.call.assert_called_once_with(text="Hello")
async def test_model_cards_and_credential_wiring(self) -> None:
- """CosyVoice model cards are discovered only by the CosyVoice class,
- and DashScope credentials expose the class."""
+ """Cards stay isolated by adapter and route through the credential."""
cosyvoice_cards = DashScopeCosyVoiceTTSModel.list_models()
+ self.assertTrue(cosyvoice_cards)
cosyvoice_names = {card.name for card in cosyvoice_cards}
- self.assertEqual(len(cosyvoice_cards), 2)
- self.assertIn("cosyvoice-v3-flash", cosyvoice_names)
- self.assertIn("cosyvoice-v3-plus", cosyvoice_names)
- self.assertTrue(all(not card.realtime for card in cosyvoice_cards))
- for card in cosyvoice_cards:
- properties = card.parameter_schema["properties"]
- self.assertIn("voice", properties)
- self.assertIn("realtime", properties)
- plus_card = next(
- card
- for card in cosyvoice_cards
- if card.name == "cosyvoice-v3-plus"
- )
- self.assertIn(
- "longanhuan",
- plus_card.parameter_schema["properties"]["voice"]["enum"],
- )
-
- qwen_names = {card.name for card in DashScopeTTSModel.list_models()}
- self.assertNotIn("cosyvoice-v3-flash", qwen_names)
- self.assertNotIn("cosyvoice-v3-plus", qwen_names)
+ qwen_names = {
+ card.name
+ for adapter in (DashScopeTTSModel, DashScopeRealtimeTTSModel)
+ for card in adapter.list_models()
+ }
+ self.assertEqual(cosyvoice_names & qwen_names, set())
credential_classes = DashScopeCredential.get_tts_model_classes()
self.assertIn(DashScopeCosyVoiceTTSModel, credential_classes)
- credential_names = {
- card.name for card in DashScopeCredential.list_tts_models()
+ credential_cards = {
+ card.name: card for card in DashScopeCredential.list_tts_models()
}
- self.assertIn("cosyvoice-v3-flash", credential_names)
- self.assertIn("cosyvoice-v3-plus", credential_names)
-
- resolved_without_realtime = _resolve_tts_class(
- credential_classes,
- "cosyvoice-v3-plus",
- )
- self.assertIs(
- resolved_without_realtime,
- DashScopeCosyVoiceTTSModel,
- )
+ for card in cosyvoice_cards:
+ with self.subTest(model=card.name):
+ self.assertEqual(credential_cards[card.name], card)
+ self.assertIs(
+ _resolve_tts_class(credential_classes, card.name),
+ DashScopeCosyVoiceTTSModel,
+ )
# ---------------------------------------------------------------------------