Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
Remove old Claude model support (#283)
Browse files* Remove Claude Opus 4.6 model support
Co-authored-by: OpenAI Codex <codex@openai.com>
* Remove Claude Opus 4.7 model support
Co-authored-by: OpenAI Codex <codex@openai.com>
* Remove Claude Sonnet 4.6 model support
Co-authored-by: OpenAI Codex <codex@openai.com>
* Remove stale Sonnet model tag test
Co-authored-by: OpenAI Codex <codex@openai.com>
---------
Co-authored-by: OpenAI Codex <codex@openai.com>
- agent/core/model_ids.py +0 -6
- agent/core/model_switcher.py +0 -4
- agent/tools/research_tool.py +6 -14
- backend/routes/agent.py +0 -7
- frontend/src/components/Chat/ChatInput.tsx +0 -8
- frontend/src/utils/model.ts +0 -2
- tests/unit/test_agent_model_gating.py +0 -1
- tests/unit/test_cli_rendering.py +4 -5
- tests/unit/test_sft_tagger.py +0 -3
agent/core/model_ids.py
CHANGED
|
@@ -2,10 +2,7 @@
|
|
| 2 |
|
| 3 |
HF_ROUTER_BASE_URL = "https://router.huggingface.co/v1"
|
| 4 |
|
| 5 |
-
CLAUDE_OPUS_46_MODEL_ID = "anthropic/claude-opus-4.6:fal-ai"
|
| 6 |
-
CLAUDE_OPUS_47_MODEL_ID = "anthropic/claude-opus-4.7:fal-ai"
|
| 7 |
CLAUDE_OPUS_48_MODEL_ID = "anthropic/claude-opus-4.8:fal-ai"
|
| 8 |
-
CLAUDE_SONNET_46_MODEL_ID = "anthropic/claude-sonnet-4-6:fal-ai"
|
| 9 |
GPT_55_MODEL_ID = "openai/gpt-5.5:fal-ai"
|
| 10 |
KIMI_K26_MODEL_ID = "moonshotai/Kimi-K2.6"
|
| 11 |
MINIMAX_M27_MODEL_ID = "MiniMaxAI/MiniMax-M2.7"
|
|
@@ -15,10 +12,7 @@ DEEPSEEK_V4_PRO_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro:deepinfra"
|
|
| 15 |
DEFAULT_MODEL_ID = CLAUDE_OPUS_48_MODEL_ID
|
| 16 |
|
| 17 |
PREMIUM_MODEL_IDS = {
|
| 18 |
-
CLAUDE_OPUS_46_MODEL_ID,
|
| 19 |
-
CLAUDE_OPUS_47_MODEL_ID,
|
| 20 |
CLAUDE_OPUS_48_MODEL_ID,
|
| 21 |
-
CLAUDE_SONNET_46_MODEL_ID,
|
| 22 |
GPT_55_MODEL_ID,
|
| 23 |
}
|
| 24 |
|
|
|
|
| 2 |
|
| 3 |
HF_ROUTER_BASE_URL = "https://router.huggingface.co/v1"
|
| 4 |
|
|
|
|
|
|
|
| 5 |
CLAUDE_OPUS_48_MODEL_ID = "anthropic/claude-opus-4.8:fal-ai"
|
|
|
|
| 6 |
GPT_55_MODEL_ID = "openai/gpt-5.5:fal-ai"
|
| 7 |
KIMI_K26_MODEL_ID = "moonshotai/Kimi-K2.6"
|
| 8 |
MINIMAX_M27_MODEL_ID = "MiniMaxAI/MiniMax-M2.7"
|
|
|
|
| 12 |
DEFAULT_MODEL_ID = CLAUDE_OPUS_48_MODEL_ID
|
| 13 |
|
| 14 |
PREMIUM_MODEL_IDS = {
|
|
|
|
|
|
|
| 15 |
CLAUDE_OPUS_48_MODEL_ID,
|
|
|
|
| 16 |
GPT_55_MODEL_ID,
|
| 17 |
}
|
| 18 |
|
agent/core/model_switcher.py
CHANGED
|
@@ -27,8 +27,6 @@ from agent.core.local_models import (
|
|
| 27 |
is_reserved_local_model_id,
|
| 28 |
)
|
| 29 |
from agent.core.model_ids import (
|
| 30 |
-
CLAUDE_OPUS_46_MODEL_ID,
|
| 31 |
-
CLAUDE_OPUS_47_MODEL_ID,
|
| 32 |
CLAUDE_OPUS_48_MODEL_ID,
|
| 33 |
GPT_55_MODEL_ID,
|
| 34 |
KIMI_K26_MODEL_ID,
|
|
@@ -42,8 +40,6 @@ from agent.core.model_ids import (
|
|
| 42 |
# policy (auto = fastest with failover).
|
| 43 |
SUGGESTED_MODELS = [
|
| 44 |
{"id": CLAUDE_OPUS_48_MODEL_ID, "label": "Claude Opus 4.8"},
|
| 45 |
-
{"id": CLAUDE_OPUS_47_MODEL_ID, "label": "Claude Opus 4.7"},
|
| 46 |
-
{"id": CLAUDE_OPUS_46_MODEL_ID, "label": "Claude Opus 4.6"},
|
| 47 |
{"id": GPT_55_MODEL_ID, "label": "GPT-5.5"},
|
| 48 |
{"id": "MiniMaxAI/MiniMax-M2.7", "label": "MiniMax M2.7"},
|
| 49 |
{"id": KIMI_K26_MODEL_ID, "label": "Kimi K2.6"},
|
|
|
|
| 27 |
is_reserved_local_model_id,
|
| 28 |
)
|
| 29 |
from agent.core.model_ids import (
|
|
|
|
|
|
|
| 30 |
CLAUDE_OPUS_48_MODEL_ID,
|
| 31 |
GPT_55_MODEL_ID,
|
| 32 |
KIMI_K26_MODEL_ID,
|
|
|
|
| 40 |
# policy (auto = fastest with failover).
|
| 41 |
SUGGESTED_MODELS = [
|
| 42 |
{"id": CLAUDE_OPUS_48_MODEL_ID, "label": "Claude Opus 4.8"},
|
|
|
|
|
|
|
| 43 |
{"id": GPT_55_MODEL_ID, "label": "GPT-5.5"},
|
| 44 |
{"id": "MiniMaxAI/MiniMax-M2.7", "label": "MiniMax M2.7"},
|
| 45 |
{"id": KIMI_K26_MODEL_ID, "label": "Kimi K2.6"},
|
agent/tools/research_tool.py
CHANGED
|
@@ -17,10 +17,7 @@ from litellm import Message, acompletion
|
|
| 17 |
from agent.core import telemetry
|
| 18 |
from agent.core.doom_loop import check_for_doom_loop
|
| 19 |
from agent.core.llm_params import _resolve_llm_params
|
| 20 |
-
from agent.core.model_ids import
|
| 21 |
-
CLAUDE_SONNET_46_MODEL_ID,
|
| 22 |
-
strip_huggingface_model_prefix,
|
| 23 |
-
)
|
| 24 |
from agent.core.session import Event
|
| 25 |
|
| 26 |
logger = logging.getLogger(__name__)
|
|
@@ -223,11 +220,8 @@ RESEARCH_TOOL_SPEC = {
|
|
| 223 |
|
| 224 |
|
| 225 |
def _get_research_model(main_model: str) -> str:
|
| 226 |
-
"""
|
| 227 |
-
|
| 228 |
-
if normalized.startswith("anthropic/claude-opus"):
|
| 229 |
-
return CLAUDE_SONNET_46_MODEL_ID
|
| 230 |
-
return normalized
|
| 231 |
|
| 232 |
|
| 233 |
async def research_handler(
|
|
@@ -252,13 +246,11 @@ async def research_handler(
|
|
| 252 |
user_content = f"Context: {context}\n\n{user_content}"
|
| 253 |
messages.append(Message(role="user", content=user_content))
|
| 254 |
|
| 255 |
-
# Use
|
| 256 |
main_model = session.config.model_name
|
| 257 |
research_model = _get_research_model(main_model)
|
| 258 |
-
# Research is a cheap sub-call — cap the main session's effort at "high"
|
| 259 |
-
#
|
| 260 |
-
# propagate to a Sonnet research model that may not accept those levels.
|
| 261 |
-
# We also haven't probed this sub-model so we don't know its ceiling.
|
| 262 |
_pref = getattr(session.config, "reasoning_effort", None)
|
| 263 |
_capped = "high" if _pref in ("max", "xhigh") else _pref
|
| 264 |
llm_params = _resolve_llm_params(
|
|
|
|
| 17 |
from agent.core import telemetry
|
| 18 |
from agent.core.doom_loop import check_for_doom_loop
|
| 19 |
from agent.core.llm_params import _resolve_llm_params
|
| 20 |
+
from agent.core.model_ids import strip_huggingface_model_prefix
|
|
|
|
|
|
|
|
|
|
| 21 |
from agent.core.session import Event
|
| 22 |
|
| 23 |
logger = logging.getLogger(__name__)
|
|
|
|
| 220 |
|
| 221 |
|
| 222 |
def _get_research_model(main_model: str) -> str:
|
| 223 |
+
"""Normalize the main model id for the research sub-call."""
|
| 224 |
+
return strip_huggingface_model_prefix(main_model) or main_model
|
|
|
|
|
|
|
|
|
|
| 225 |
|
| 226 |
|
| 227 |
async def research_handler(
|
|
|
|
| 246 |
user_content = f"Context: {context}\n\n{user_content}"
|
| 247 |
messages.append(Message(role="user", content=user_content))
|
| 248 |
|
| 249 |
+
# Use the normalized router model for research
|
| 250 |
main_model = session.config.model_name
|
| 251 |
research_model = _get_research_model(main_model)
|
| 252 |
+
# Research is a cheap sub-call — cap the main session's effort at "high".
|
| 253 |
+
# We also haven't probed this sub-call's model so we don't know its ceiling.
|
|
|
|
|
|
|
| 254 |
_pref = getattr(session.config, "reasoning_effort", None)
|
| 255 |
_capped = "high" if _pref in ("max", "xhigh") else _pref
|
| 256 |
llm_params = _resolve_llm_params(
|
backend/routes/agent.py
CHANGED
|
@@ -55,7 +55,6 @@ from agent.core.hf_access import get_jobs_access
|
|
| 55 |
from agent.core.hf_tokens import resolve_hf_request_token
|
| 56 |
from agent.core.llm_params import _resolve_llm_params
|
| 57 |
from agent.core.model_ids import (
|
| 58 |
-
CLAUDE_OPUS_46_MODEL_ID,
|
| 59 |
DEEPSEEK_V4_PRO_MODEL_ID,
|
| 60 |
DEFAULT_MODEL_ID,
|
| 61 |
GLM_51_MODEL_ID,
|
|
@@ -85,12 +84,6 @@ def _available_models() -> list[dict[str, Any]]:
|
|
| 85 |
"tier": "pro",
|
| 86 |
"recommended": True,
|
| 87 |
},
|
| 88 |
-
{
|
| 89 |
-
"id": CLAUDE_OPUS_46_MODEL_ID,
|
| 90 |
-
"label": "Claude Opus 4.6",
|
| 91 |
-
"provider": "huggingface",
|
| 92 |
-
"tier": "pro",
|
| 93 |
-
},
|
| 94 |
{
|
| 95 |
"id": DEFAULT_GPT_MODEL_ID,
|
| 96 |
"label": "GPT-5.5",
|
|
|
|
| 55 |
from agent.core.hf_tokens import resolve_hf_request_token
|
| 56 |
from agent.core.llm_params import _resolve_llm_params
|
| 57 |
from agent.core.model_ids import (
|
|
|
|
| 58 |
DEEPSEEK_V4_PRO_MODEL_ID,
|
| 59 |
DEFAULT_MODEL_ID,
|
| 60 |
GLM_51_MODEL_ID,
|
|
|
|
| 84 |
"tier": "pro",
|
| 85 |
"recommended": True,
|
| 86 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
{
|
| 88 |
"id": DEFAULT_GPT_MODEL_ID,
|
| 89 |
"label": "GPT-5.5",
|
frontend/src/components/Chat/ChatInput.tsx
CHANGED
|
@@ -25,7 +25,6 @@ import JobsUpgradeDialog from '@/components/JobsUpgradeDialog';
|
|
| 25 |
import { useAgentStore } from '@/store/agentStore';
|
| 26 |
import { useSessionStore } from '@/store/sessionStore';
|
| 27 |
import {
|
| 28 |
-
CLAUDE_OPUS_46_MODEL_PATH,
|
| 29 |
CLAUDE_MODEL_PATH,
|
| 30 |
GPT_55_MODEL_PATH,
|
| 31 |
isClaudePath,
|
|
@@ -56,13 +55,6 @@ const DEFAULT_MODEL_OPTIONS: ModelOption[] = [
|
|
| 56 |
avatarUrl: getHfAvatarUrl(CLAUDE_MODEL_PATH),
|
| 57 |
recommended: true,
|
| 58 |
},
|
| 59 |
-
{
|
| 60 |
-
id: 'claude-opus-4-6',
|
| 61 |
-
name: 'Claude Opus 4.6',
|
| 62 |
-
description: 'Hugging Face',
|
| 63 |
-
modelPath: CLAUDE_OPUS_46_MODEL_PATH,
|
| 64 |
-
avatarUrl: getHfAvatarUrl(CLAUDE_OPUS_46_MODEL_PATH),
|
| 65 |
-
},
|
| 66 |
{
|
| 67 |
id: 'gpt-5.5',
|
| 68 |
name: 'GPT-5.5',
|
|
|
|
| 25 |
import { useAgentStore } from '@/store/agentStore';
|
| 26 |
import { useSessionStore } from '@/store/sessionStore';
|
| 27 |
import {
|
|
|
|
| 28 |
CLAUDE_MODEL_PATH,
|
| 29 |
GPT_55_MODEL_PATH,
|
| 30 |
isClaudePath,
|
|
|
|
| 55 |
avatarUrl: getHfAvatarUrl(CLAUDE_MODEL_PATH),
|
| 56 |
recommended: true,
|
| 57 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
{
|
| 59 |
id: 'gpt-5.5',
|
| 60 |
name: 'GPT-5.5',
|
frontend/src/utils/model.ts
CHANGED
|
@@ -6,13 +6,11 @@
|
|
| 6 |
* AVAILABLE_MODELS in backend/routes/agent.py.
|
| 7 |
*/
|
| 8 |
|
| 9 |
-
export const CLAUDE_OPUS_46_MODEL_PATH = 'anthropic/claude-opus-4.6:fal-ai';
|
| 10 |
export const CLAUDE_OPUS_48_MODEL_PATH = 'anthropic/claude-opus-4.8:fal-ai';
|
| 11 |
export const CLAUDE_MODEL_PATH = CLAUDE_OPUS_48_MODEL_PATH;
|
| 12 |
export const GPT_55_MODEL_PATH = 'openai/gpt-5.5:fal-ai';
|
| 13 |
|
| 14 |
const PREMIUM_MODEL_PATHS = new Set([
|
| 15 |
-
CLAUDE_OPUS_46_MODEL_PATH,
|
| 16 |
CLAUDE_OPUS_48_MODEL_PATH,
|
| 17 |
GPT_55_MODEL_PATH,
|
| 18 |
]);
|
|
|
|
| 6 |
* AVAILABLE_MODELS in backend/routes/agent.py.
|
| 7 |
*/
|
| 8 |
|
|
|
|
| 9 |
export const CLAUDE_OPUS_48_MODEL_PATH = 'anthropic/claude-opus-4.8:fal-ai';
|
| 10 |
export const CLAUDE_MODEL_PATH = CLAUDE_OPUS_48_MODEL_PATH;
|
| 11 |
export const GPT_55_MODEL_PATH = 'openai/gpt-5.5:fal-ai';
|
| 12 |
|
| 13 |
const PREMIUM_MODEL_PATHS = new Set([
|
|
|
|
| 14 |
CLAUDE_OPUS_48_MODEL_PATH,
|
| 15 |
GPT_55_MODEL_PATH,
|
| 16 |
]);
|
tests/unit/test_agent_model_gating.py
CHANGED
|
@@ -33,7 +33,6 @@ def _premium_session(model: str = agent.DEFAULT_PREMIUM_MODEL_ID):
|
|
| 33 |
|
| 34 |
def test_premium_model_predicate_uses_router_ids_only():
|
| 35 |
assert agent._is_premium_model(agent.DEFAULT_PREMIUM_MODEL_ID)
|
| 36 |
-
assert agent._is_premium_model(agent.CLAUDE_OPUS_46_MODEL_ID)
|
| 37 |
assert agent._is_premium_model(agent.DEFAULT_GPT_MODEL_ID)
|
| 38 |
assert agent._is_user_billed(agent.DEFAULT_PREMIUM_MODEL_ID)
|
| 39 |
assert not agent._is_premium_model("moonshotai/Kimi-K2.6")
|
|
|
|
| 33 |
|
| 34 |
def test_premium_model_predicate_uses_router_ids_only():
|
| 35 |
assert agent._is_premium_model(agent.DEFAULT_PREMIUM_MODEL_ID)
|
|
|
|
| 36 |
assert agent._is_premium_model(agent.DEFAULT_GPT_MODEL_ID)
|
| 37 |
assert agent._is_user_billed(agent.DEFAULT_PREMIUM_MODEL_ID)
|
| 38 |
assert not agent._is_premium_model("moonshotai/Kimi-K2.6")
|
tests/unit/test_cli_rendering.py
CHANGED
|
@@ -13,10 +13,10 @@ from agent.tools.research_tool import _get_research_model
|
|
| 13 |
from agent.utils import terminal_display
|
| 14 |
|
| 15 |
|
| 16 |
-
def
|
| 17 |
assert (
|
| 18 |
_get_research_model("anthropic/claude-opus-4.8:fal-ai")
|
| 19 |
-
== "anthropic/claude-
|
| 20 |
)
|
| 21 |
|
| 22 |
|
|
@@ -26,13 +26,12 @@ def test_non_anthropic_research_model_is_unchanged():
|
|
| 26 |
|
| 27 |
def test_huggingface_prefix_research_model_strips_prefix():
|
| 28 |
assert (
|
| 29 |
-
_get_research_model("huggingface/anthropic/claude-opus-4.
|
| 30 |
-
== "anthropic/claude-
|
| 31 |
)
|
| 32 |
|
| 33 |
|
| 34 |
def test_hf_router_openai_research_model_is_unchanged():
|
| 35 |
-
# No Sonnet equivalent for GPT — research stays on the same router model.
|
| 36 |
assert (
|
| 37 |
_get_research_model("huggingface/openai/gpt-5.5:fal-ai")
|
| 38 |
== "openai/gpt-5.5:fal-ai"
|
|
|
|
| 13 |
from agent.utils import terminal_display
|
| 14 |
|
| 15 |
|
| 16 |
+
def test_router_anthropic_research_model_is_unchanged():
|
| 17 |
assert (
|
| 18 |
_get_research_model("anthropic/claude-opus-4.8:fal-ai")
|
| 19 |
+
== "anthropic/claude-opus-4.8:fal-ai"
|
| 20 |
)
|
| 21 |
|
| 22 |
|
|
|
|
| 26 |
|
| 27 |
def test_huggingface_prefix_research_model_strips_prefix():
|
| 28 |
assert (
|
| 29 |
+
_get_research_model("huggingface/anthropic/claude-opus-4.8:fal-ai")
|
| 30 |
+
== "anthropic/claude-opus-4.8:fal-ai"
|
| 31 |
)
|
| 32 |
|
| 33 |
|
| 34 |
def test_hf_router_openai_research_model_is_unchanged():
|
|
|
|
| 35 |
assert (
|
| 36 |
_get_research_model("huggingface/openai/gpt-5.5:fal-ai")
|
| 37 |
== "openai/gpt-5.5:fal-ai"
|
tests/unit/test_sft_tagger.py
CHANGED
|
@@ -21,9 +21,6 @@ def _traj(events=None, messages=None, model="anthropic/claude-opus-4.8:fal-ai"):
|
|
| 21 |
|
| 22 |
def test_model_family():
|
| 23 |
assert "model:opus" in tag_session(_traj(model="anthropic/claude-opus-4.8:fal-ai"))
|
| 24 |
-
assert "model:sonnet" in tag_session(
|
| 25 |
-
_traj(model="anthropic/claude-sonnet-4-6:fal-ai")
|
| 26 |
-
)
|
| 27 |
assert "model:kimi" in tag_session(_traj(model="moonshotai/Kimi-K2.6"))
|
| 28 |
assert "model:other" in tag_session(_traj(model="unknown-model-xyz"))
|
| 29 |
|
|
|
|
| 21 |
|
| 22 |
def test_model_family():
|
| 23 |
assert "model:opus" in tag_session(_traj(model="anthropic/claude-opus-4.8:fal-ai"))
|
|
|
|
|
|
|
|
|
|
| 24 |
assert "model:kimi" in tag_session(_traj(model="moonshotai/Kimi-K2.6"))
|
| 25 |
assert "model:other" in tag_session(_traj(model="unknown-model-xyz"))
|
| 26 |
|