run_python — Sandbox Reference

Available helpers

These are the ONLY callable names. Do not use tool names (e.g. fetch_latest_feed, search_openrouter_models — those are tool names, not Python functions).

Helper	Returns
fetch_benchmarks(query='', limit=50)	list of dicts: name, provider, quality, coding, math, speed_tps, input_per_1m, output_per_1m, context_k
fetch_news(days_back=1, include_hype=False, query='', limit=50, since=None, until=None)	list of dicts: title, source, date, url, summary
fetch_openrouter(query='', limit=50)	list of dicts: id, name, provider, context_k, input_per_1m, output_per_1m
fetch_image_models(limit=30)	list of dicts: name, provider, elo, cost
fetch_video_models(limit=20)	list of dicts: name, provider, elo

Sandbox rules

• No import statements — __import__ is blocked. Standard libs (json, re, math, datetime, collections) are already available.
• No network access — only the helpers above can fetch data.
• Use print() for output — return values are ignored. Print only what matters.
• Keys may be missing — use .get('key') not ['key'] to avoid KeyError on sparse data.

Common patterns

Top models for coding:

models = fetch_benchmarks(limit=100)
coding = [m for m in models if m.get('coding')]
for m in sorted(coding, key=lambda x: x['coding'], reverse=True)[:10]:
    print(m['name'], 'coding=' + str(m['coding']), 'general=' + str(m.get('quality', '?')))

Price/quality ratio:

models = fetch_benchmarks(query='anthropic', limit=50)
for m in models:
    q = m.get('quality') or 0
    cost = m.get('output_per_1m') or 0
    if q and cost:
        print(m['name'], 'ratio=' + str(round(q / cost, 1)), 'q=' + str(q), '$' + str(cost) + '/1M')

Date-range news:

items = fetch_news(days_back=30, since='2026-02-01', until='2026-02-10', limit=100)
print(len(items), 'items in range')
for it in items[:10]:
    print('[' + (it.get('date') or '')[:10] + ']', it.get('title', '?'))

Cross-tool join (benchmarks × OpenRouter):

bench = fetch_benchmarks(limit=30)
or_models = fetch_openrouter(query='openai', limit=50)
or_ids = {m.get('id', '').lower() for m in or_models}
for b in bench[:10]:
    available = any(b['name'].lower() in oid for oid in or_ids)
    print(('[OR]' if available else '[  ]'), b['name'], 'q=' + str(b.get('quality', '?')))

What to use for each question

Question type	Use coding	Use quality	Use math
Best for code generation	✓	–	–
Best overall / general	–	✓	–
Best for math/science	–	–	✓
Best value (general)	–	✓	–