Claude Opus 5.5 vs GPT-5.6 Luna: pricing and benchmarks
GPT-5.6 Luna is 95% cheaper on input tokens ($0.20 vs $4.00 per 1M). GPT-5.6 Luna is 94% cheaper on output tokens ($1.20 vs $20.00 per 1M). On our document Q&A workload (50,000 questions a month), GPT-5.6 Luna costs $116.00 a month versus $2,200 for Claude Opus 5.5, 95% less. Claude Opus 5.5 scores 166.8 on the Epoch Capabilities Index vs 156.3 for GPT-5.6 Luna.
Which should you choose?
Claude Opus 5.5 scores 166.8 on the Epoch Capabilities Index vs 156.3 for GPT-5.6 Luna.
Claude Opus 5.5 scores 91.2% on FrontierMath (Tiers 1–3) vs 82.1% for GPT-5.6 Luna.
GPT-5.6 Luna costs $0.45 per 1M tokens (3:1 input-to-output mix) vs $8.00 for Claude Opus 5.5, 94% less.
A trade-off: Claude Opus 5.5 is 10.5 ECI points more capable but costs 17.8x as much as GPT-5.6 Luna.
Pricing and specs
| Claude Opus 5.5 | GPT-5.6 Luna | |
|---|---|---|
| Provider | Anthropic | OpenAI |
| API model ID | claude-opus-5-5 | gpt-5.6-luna |
| Input / 1M tokens | $4.00 | $0.20 |
| Cached input / 1M tokens | $0.20 | $0.02 |
| Output / 1M tokens | $20.00 | $1.20 |
| Long-context rates | — | $0.40 in / $1.80 out on long-context requests |
| Context window | 1M tokens | 1.05M tokens |
| Max output | 128K tokens | 128K tokens |
| Batch discount | 50% off | 50% off |
Benchmarks
| Claude Opus 5.5 | GPT-5.6 Luna | |
|---|---|---|
| Epoch Capabilities Index | 166.8 | 156.3 |
| GPQA Diamond· Science reasoning | 90.6% | 91.6% |
| FrontierMath (Tiers 1–3)· Advanced math | 91.2% | 82.1% |
| SimpleQA Verified· Factual accuracy | 72.2% | 41.0% |
| ARC-AGI-2· Abstract reasoning | 93.3% | 59.5% |
| DeepSWE· Coding | — | 67.2% |
| APEX-Agents· Agentic work | 73.5% | 43.0% |
Source: Epoch AI, best recorded result per model (CC BY 4.0). A dash means no published score. See all benchmarks
Monthly cost for real workloads
| Workload | Claude Opus 5.5 | GPT-5.6 Luna | Difference |
|---|---|---|---|
Support chatbot 100,000 replies a month, ~1,500 input tokens (system prompt + history) and ~400 output tokens each. | $1,400 | $78.00 | GPT-5.6 Luna 94% cheaper |
Document Q&A (RAG) 50,000 questions a month with ~8,000 tokens of retrieved context and ~600 output tokens each. | $2,200 | $116.00 | GPT-5.6 Luna 95% cheaper |
Coding agent 10,000 agent steps a month, ~40,000 input tokens each (70% read from the prompt cache) and ~2,000 output tokens. | $936.00 | $53.60 | GPT-5.6 Luna 94% cheaper |
Cached chatbot The same 100,000-reply support chatbot, but 60% of each prompt (system prompt + conversation history) is served from the prompt cache. | $1,058 | $61.80 | GPT-5.6 Luna 94% cheaper |
Meeting notes 10,000 meetings a month summarized from ~12,000-token transcripts, ~1,000 output tokens each. | $680.00 | $36.00 | GPT-5.6 Luna 95% cheaper |
Bulk translation 200,000 short translations a month, ~800 input and ~800 output tokens each. | $3,840 | $224.00 | GPT-5.6 Luna 94% cheaper |
Content moderation 1,000,000 classifications a month, ~400 input tokens and ~80 output tokens each. | $3,200 | $176.00 | GPT-5.6 Luna 95% cheaper |
Article drafting 300 long-form articles a month written from a ~4,000-token brief and outline, ~5,000 output tokens each. | $34.80 | $2.04 | GPT-5.6 Luna 94% cheaper |
Code review bot 30,000 pull-request reviews a month, ~5,000 input tokens each (30% cached repository context) and ~700 output tokens. | $849.00 | $47.10 | GPT-5.6 Luna 94% cheaper |
Invoice extraction 100,000 documents a month parsed into structured JSON, ~1,200 input and ~250 output tokens each. | $980.00 | $54.00 | GPT-5.6 Luna 94% cheaper |
Estimates use standard list prices and ignore cache-write surcharges. Different tokenizers can count the same text differently, so test with your own prompts.
Claude Opus 5.5 vs GPT-5.6 Luna cost calculator
- $116.00/mo$0.0023 / req
- Claude Opus 5.5Anthropic$2,200/mo$0.04 / req
Standard-tier list prices. Long-context rates apply automatically where the provider publishes a threshold. Excludes cache-write surcharges, taxes, batch and volume discounts.
Claude Opus 5.5 pricing notes
Anthropic’s recommended default model for most workloads, with adaptive thinking always on.
- Cache reads cost 5% of the input price, against 10% on Claude Opus 5 and Sonnet 5.
- Cache writes cost $5 (5-minute cache) or $8 (1-hour cache) per 1M tokens.
- Fast mode costs $8 input / $40 output per 1M tokens.
- US-only inference (inference_geo: "us") costs 1.1x.
GPT-5.6 Luna pricing notes
The GPT-5.6 model optimized for cost-sensitive, high-volume workloads.
- Cache writes are billed at $0.25 per 1M tokens.
- Fast mode costs 2x the standard rate; Flex costs half.
Frequently asked questions
- Is Claude Opus 5.5 cheaper than GPT-5.6 Luna?
- GPT-5.6 Luna is 95% cheaper on input tokens ($0.20 vs $4.00 per 1M). GPT-5.6 Luna is 94% cheaper on output tokens ($1.20 vs $20.00 per 1M). On our document Q&A workload (50,000 questions a month), GPT-5.6 Luna costs $116.00 a month versus $2,200 for Claude Opus 5.5, 95% less.
- Which is more capable, Claude Opus 5.5 or GPT-5.6 Luna?
- Claude Opus 5.5 scores 166.8 on the Epoch Capabilities Index vs 156.3 for GPT-5.6 Luna.
- How much does Claude Opus 5.5 cost per 1M tokens?
- Claude Opus 5.5 costs $4.00 per 1M input tokens and $20.00 per 1M output tokens, and $0.20 per 1M cached input tokens on Anthropic’s standard tier.
- How much does GPT-5.6 Luna cost per 1M tokens?
- GPT-5.6 Luna costs $0.20 per 1M input tokens and $1.20 per 1M output tokens, and $0.02 per 1M cached input tokens on OpenAI’s standard tier.
- Which has the larger context window, Claude Opus 5.5 or GPT-5.6 Luna?
- GPT-5.6 Luna has the larger context window: 1.05M tokens versus 1M for Claude Opus 5.5.