Verdict
Comparing model costs by per-token list price alone is misleading. GPT-5.5’s output price is 3x Gemini 2.5 Pro’s, but the total cost to run the full Artificial Analysis Intelligence Index is $3,357 vs $861 — less than 4x the gap, because GPT-5.5 uses far fewer tokens.
True cost ranking (low to high): Tencent Hy3 Preview (free) > DeepSeek V4 Pro ($1,071) > Gemini 2.5 Pro ($861) > GPT-5.5 medium ($1,199) > GPT-5.5 high ($2,159) > GPT-5.5 xhigh ($3,357) > Claude Sonnet 4.6 ($3,959) > Claude Opus 4.7 ($4,811).
Test Dimensions
List Price
| Model | Input ($/MTok) | Output ($/MTok) |
|---|---|---|
| GPT-5.5 | $5 | $30 |
| Claude Opus 4.7 | $5 | $25 |
| Claude Sonnet 4.6 | $3 | $15 |
| Gemini 2.5 Pro | $1.25 | $10 |
| DeepSeek V4 | $0.3 | $3.48 |
| Tencent Hy3 Preview | $0 | $0 |
Actual Task Cost
Full Artificial Analysis Intelligence Index (10 standardized evaluations):
- Claude Opus 4.7: $4,811
- Claude Sonnet 4.6: $3,959
- GPT-5.5 (xhigh): $3,357
- GPT-5.4 (xhigh): $2,851
- GPT-5.5 (high): $2,159
- DeepSeek V4 Pro: $1,071
- GPT-5.5 (medium): $1,199
- Gemini 2.5 Pro: $861
Key finding: GPT-5.5, despite the highest list price, costs 30% less than Claude Opus 4.7 for the same benchmark suite — token efficiency offsets the unit price disadvantage.
GitHub Copilot Multipliers
Latest model multipliers:
- Opus 4.6 / Sonnet 4.6: 9x
- Opus 4.5 / Sonnet 4.5: 6x (Sonnet), 5x (Opus)
- Opus 4.7: 3.6x
- Gemini 3 Pro / 3.1 Pro: 6x
- GPT 5.1: 4x
Scenario Cost Estimates
1,000 customer service chats/day (~2K tokens each):
- Gemini 2.5 Pro: ~$2.5/day
- DeepSeek V4: ~$7.6/day
- GPT-5.5 (medium): ~$10/day
- Claude Opus 4.7: ~$25/day
50 complex code reviews/day (~20K tokens each):
- Gemini 2.5 Pro: ~$12.5/day
- DeepSeek V4 Pro: ~$18/day
- GPT-5.5 (high): ~$35/day
- Claude Sonnet 4.6: ~$45/day
Recommendations
Cost-first (simple tasks): Gemini 2.5 Pro.
Cost-performance balance: GPT-5.5 (medium/high quality tiers).
极限 quality: GPT-5.5 (xhigh) — 30% cheaper than Opus 4.7.
Offline / self-hosted: DeepSeek V4 or Qwen 3.6-27B.
GitHub Copilot users: Watch multiplier changes — Opus models at 3.6x-9x.