NVIDIA inference API advertises 1M context for Opus 4.6 but backend rejects >200K tokens (both aws/ and us/aws/ model IDs)

test_nvidia_opus_context.py

#!/usr/bin/env python3
"""Opus 4.6 advertises 1M context on inference.nvidia.com but rejects >200K."""

import json, os, sys, urllib.request

API_KEY = os.environ.get("NVIDIA_API_KEY") or os.environ.get("NVIDIA_INFERENCE_KEY")
if not API_KEY:
    sys.exit("Set NVIDIA_API_KEY env var (from https://inference.nvidia.com)")

# ~950K tokens of filler (well under advertised 1M limit)
# 3.8M chars / ~4 chars per token ≈ 950K tokens (API reported 928,958 actual)
filler = ("The quick brown fox jumps over the lazy dog. " * 85_000)[:3_800_000]

payload = json.dumps({
    "model": "aws/anthropic/bedrock-claude-opus-4-6",
    "messages": [
        {"role": "user", "content": filler + "\n\nReply with 'OK'"},
    ],
    "max_tokens": 10,
}).encode()

print(f"Sending ~950K tokens to Opus 4.6 ({len(payload)/1e6:.1f} MB)...")

req = urllib.request.Request(
    "https://inference-api.nvidia.com/v1/chat/completions",
    data=payload,
    headers={"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"},
)

try:
    with urllib.request.urlopen(req, timeout=300) as r:
        print("SUCCESS:", json.loads(r.read())["choices"][0]["message"]["content"])
except urllib.error.HTTPError as e:
    print("FAILED:", json.loads(e.read())["error"]["message"])