V2 Documentation — GPT-5 Mini Direct

Overview

V2 makes a single API call to gpt-5-mini-2025-08-07 with the full consumer prompt. No web search — the model uses its training knowledge for pricing, competitors, and company facts. Surprisingly thorough for most popular services.

Attribute	Value
Model	gpt-5-mini-2025-08-07
Web searches	None
Avg article length	1,700–2,100 words
Avg quality score	9/10
Cost per article	€0.009–0.013
Avg generation time	60–100 seconds
Parallelisable	Yes — 5 workers simultaneously
Cost × 1,000 articles	≈ €9–13

Architecture

┌─────────────────────────────────────────────────────┐
│  SERVICE DATA (name, category, notes, keywords…)    │
└────────────────────────┬────────────────────────────┘
                         │
         ┌──────────────▼──────────────────┐
         │         SINGLE API CALL          │
         │                                  │
         │  Consumer Prompt Template        │  gpt-5-mini
         │  + Service context block         │  max_completion_tokens=8000
         │  + 14-section structure          │  streaming=true
         │  Model uses training knowledge   │
         │  for pricing + competitors       │
         └──────────────┬───────────────────┘
                        │
                  FINAL ARTICLE
                  (HTML, ~1,900 words, 14 H2, 4 tables)

Prompt Context Injected

V2 injects a "Service Context" block instead of live research:

══ SERVICE CONTEXT ══
Cancellation methods: phone, email, account-online
Notes: {service notes from JSON, first 200 chars}
Keywords: cancel netflix, how to cancel netflix australia, ...

The model draws on its training data for all other facts (company HQ, CEO, pricing history, competitors).

When to use V2

Large-scale generation of 100–10,000 articles where cost matters
Services where pricing stability is low (model knowledge may be slightly outdated)
Initial content generation before manual review
Categories with well-known services (streaming, software, SaaS)

When NOT to use V2

Niche or local services the model may not know well
Services with very recent pricing changes
When you need verified, sourced competitor data

Cost Breakdown

Component	Tokens (avg)	Cost (USD)	Cost (EUR)
Prompt tokens	~1,900	$0.000475	€0.00044
Completion tokens	~5,700	$0.01140	€0.01049
TOTAL per article	~7,600	~$0.012	~€0.011
× 20 articles	~152,000	$0.24	€0.22
× 1,000 articles	~7.6M	$11.87	€10.92

Model pricing: Input $0.25/1M tokens, Output $2.00/1M tokens.

Python Script Reference

# Run V2 batch (all 20 sample services, 5 parallel workers)
python scripts/02_generate_v2.py

# Output: Billoff/data/results_v2.json

# Run 3-method comparison on 1 service
python scripts/test_compare_3methods.py --service "Netflix"

Parallelisation

# V2 runs with 5 concurrent workers (no rate limit concern for direct completions)
from concurrent.futures import ThreadPoolExecutor
MAX_WORKERS = 5

with ThreadPoolExecutor(max_workers=MAX_WORKERS) as ex:
    futures = {ex.submit(process_service, svc): svc for svc in services}
    for future in as_completed(futures):
        results.append(future.result())

Quality Results (Ocado test)

Metric	Result	vs V1
Word count	1,927	−192 (−9%)
Tables	4	Same
H2 sections	14	Same
H3 sub-sections	40	+1 (better!)
Company fact box	✅	Same
FAQ	✅	Same
Quality score	9/10	−1
Cost	€0.011	10× cheaper
Time	95s	−31% faster

Pros & Cons

✅ Pros	❌ Cons
10× cheaper than V1	No real-time pricing data
Parallelisable (5+ workers)	Competitors may not have current prices
Consistently good structure (14 H2)	Company info may be outdated (training cutoff)
9/10 quality score in tests	Less reliable for niche/local services
Best cost/quality ratio for scale	No verified source citations
Works offline (no web dependency)	May hallucinate pricing for obscure services

⚡ GPT-5 Mini Direct (V2)