Update LLM Model List

cat > /tmp/check_agenta_models.py << 'SCRIPT'
# /// script
# requires-python = ">=3.11"
# dependencies = ["litellm"]
# ///
import litellm, sys

# paste supported_llm_models here or import it
from agenta.sdk.utils.assets import supported_llm_models

mc = set(litellm.model_cost.keys())

def exists(m):
    if m in mc: return True
    if "/" in m and m.split("/", 1)[1] in mc: return True
    return False

fails = []
for provider, models in supported_llm_models.items():
    for model in models:
        if not exists(model):
            fails.append((provider, model))

total = sum(len(v) for v in supported_llm_models.values())
print(f"Total models checked: {total}")
if fails:
    for p, m in fails:
        print(f"  MISSING [{p}] {m}")
    sys.exit(1)
else:
    print("All models valid ✓")
SCRIPT
uvx --with litellm python /tmp/check_agenta_models.py 2>/dev/null

pytest sdks/python/oss/tests/pytest/unit/test_supported_llm_models.py -v

cat > /tmp/find_missing.py << 'SCRIPT'
# /// script
# requires-python = ">=3.11"
# dependencies = ["litellm"]
# ///
import litellm, re

AGENTA_ANTHROPIC = set()   # fill from assets.py (bare names, no prefix)
AGENTA_OPENAI    = set()   # fill from assets.py
AGENTA_GEMINI    = set()   # fill from assets.py (with gemini/ prefix)

mc = set(litellm.model_cost.keys())

NOISE = [
    "audio","tts","speech","whisper","transcri","realtime","diarize",
    "dall-e","image","video","veo","embed","moderat","search",
    "babbage","davinci","ada","instruct","codex","computer-use",
    "robotics","learnlm","gemma","live","v1:0",
]
KEEP = {"gpt-4o","gpt-4o-mini"}
DATED = re.compile(r"-\d{4}-\d{2}-\d{2}$")
EXP   = re.compile(r"exp-\d{4}|\d{2}-\d{2}$")

def noise(m):
    if m in KEEP: return False
    return any(kw in m.lower() for kw in NOISE)

def dated(m):
    return bool(DATED.search(m)) or bool(EXP.search(m))

def report(label, candidates, known, prefix=""):
    print(f"\n=== {label} ===")
    for m in sorted(candidates):
        bare = m[len(prefix):] if prefix else m
        if bare in known or m in known: continue
        tag = "[dated/exp]" if dated(m) else "[alias]" if m.endswith("-latest") else "*** MISSING ***"
        print(f"  {m}  {tag}")

# Anthropic
report("ANTHROPIC", [m for m in mc if m.startswith("claude-") and not noise(m)],
       AGENTA_ANTHROPIC)

# OpenAI (no slash, starts with gpt- / o1 / o3 / o4)
OAI = [m for m in mc if any(m.startswith(p) for p in ("gpt-","o1","o3","o4","chatgpt"))
       and "/" not in m and not noise(m)]
report("OPENAI", OAI, AGENTA_OPENAI)

# Gemini
report("GEMINI", [m for m in mc if m.startswith("gemini/") and not noise(m)],
       AGENTA_GEMINI, prefix="gemini/")
SCRIPT
uvx --with litellm python /tmp/find_missing.py 2>/dev/null

# Format + lint
uvx --from ruff==0.14.0 ruff format sdks/python/agenta/sdk/utils/assets.py
uvx --from ruff==0.14.0 ruff check --fix sdks/python/agenta/sdk/utils/assets.py

# Validate all models against litellm (no agenta install needed)
uvx --with litellm python /tmp/check_agenta_models.py 2>/dev/null