Haiku 5.5 · over 100K
- Input: $0.50 / MTok
- Output: $2.50 / MTok
- Cache read: $0.05 / MTok
- 5-minute cache write: $0.625 / MTok
MODEL COMPARISON · VERIFIED OCTOBER 8, 2026
Haiku 5.5
vs GPT-6 Luna
Two low-cost models with the same short-context token rates—but different long-context thresholds, reasoning controls and reported benchmark results.
At short context, price is a tie. The real decision is workload fit.
Both models list $0.10 input and $0.50 output per million tokens, plus $0.01 cached input and $0.125 cache writes. Haiku 5.5 has stronger results on the selected benchmarks Anthropic published; GPT-6 Luna offers a 1.05M context window and does not enter its higher-price tier until prompts exceed 272K input tokens.
01 / PRICE
The first tier is identical.
Standard API prices in USD per one million tokens. Provider and processing-tier prices can differ.
| Token category | Haiku 5.5 ≤100K prompt | GPT-6 Luna ≤272K input |
|---|---|---|
| Input | $0.10 | $0.10 |
| Output | $0.50 | $0.50 |
| Cached input / cache read | $0.01 | $0.01 |
| Cache write | $0.125 · 5 min | $0.125 |
| Batch processing | 50% off input/output | 50% of Standard rates |
Equal rate cards do not guarantee equal bills. Tokenizers, reasoning tokens, output length, retries and tool calls can change the cost per completed task.
02 / LONG CONTEXT
The thresholds create the biggest pricing difference.
Each provider applies its higher tier to the full qualifying request.
GPT-6 Luna · over 272K
- Input: $0.20 / MTok
- Output: $0.75 / MTok
- Cached input: $0.02 / MTok
- Cache write: $0.25 / MTok
Between 100K and 272K prompt tokens, Haiku 5.5 is already in its higher tier while GPT-6 Luna remains at its short-context rate. This is a rate-card comparison only; task quality and token use still matter.
03 / WORKED EXAMPLE
At 200K input and 50K output, the list-price gap opens.
One standard request, no cache hits or tool fees.
Haiku 5.5: 200K × $0.50/MTok + 50K × $2.50/MTok.
GPT-6 Luna: 200K × $0.10/MTok + 50K × $0.50/MTok.
Haiku’s list cost in this specific token-count example. This is not a quality-adjusted comparison.
04 / BENCHMARKS
Anthropic reports Haiku ahead on these tests.
These figures come from Anthropic’s Haiku 5.5 announcement. Vendor-reported benchmark comparisons should be validated on your own tasks.
| Benchmark | Haiku 5.5 | GPT-6 Luna |
|---|---|---|
| GDPval-AA v2.1 | 1620 | 1437 |
| AA-Briefcase v1.1 | 1578 | 1336 |
| OSWorld 2.1 · offline subset | 72.4% | 48.9% |
| Terminal-Bench 4.0 | 39.2% | 16.4% |
| FrontierCode 1.1 · main | 46.4% | 42.4% |
| Chartography · no tools | 46.4% | 29.1% |
05 / DECISION
Which model should you test first?
Both target high-volume, cost-sensitive work.
Start with Haiku 5.5
- Anthropic-native systems and Claude model routing
- Short, narrow tasks such as classification and compaction
- The published benchmark mix matches your workload
- Your prompts normally remain under 100K
Start with GPT-6 Luna
- You need a documented 1.05M context window
- Prompts commonly fall between 100K and 272K
- You need OpenAI Responses API tools or reasoning controls
- Your existing infrastructure is built around OpenAI APIs
best model = lowest total cost per accepted resultMeasure success rate, p50/p95 latency, input and output tokens, retries, cache hit rate and any paid tool calls on the same evaluation set.
06 / FAQ
Quick answers
Is Haiku 5.5 cheaper than GPT-6 Luna?
Not in their short-context tiers: both list $0.10 input, $0.50 output, $0.01 cached input and $0.125 cache writes per million tokens. Their higher tiers begin at different thresholds.
Which model has the larger context window?
OpenAI documents a 1,050,000-token context window for GPT-6 Luna. Haiku 5.5’s 100K figure on this page is a pricing threshold, not necessarily its maximum context limit.
When does Luna’s long-context price apply?
OpenAI says prompts with more than 272K input tokens use 2× input and cache rates and 1.5× output pricing for the full request.
Are the benchmark results independently verified?
The comparison table reproduces results published by Anthropic. Treat them as vendor-reported data and run a representative evaluation before selecting a model.
What are the model IDs?
Anthropic lists claude-haiku-5-5; OpenAI lists gpt-6-luna.
Method and sources. Prices, model IDs and specifications were checked against official documentation on October 8, 2026. Benchmark figures are vendor-reported by Anthropic. This site is independent and is not affiliated with Anthropic or OpenAI. Anthropic source · OpenAI model page · OpenAI pricing.