Below 100,000 input tokens, Claude Haiku 5.5 and GPT-6 Luna have the same listed API rates: $0.10 per million input tokens and $0.50 per million output tokens.
Above 100K, Haiku's rates rise to $0.50 input and $2.50 output. Luna keeps its lower rates through 272K input tokens. This price difference is the main point in Kingy's comparison.
Which tasks fit each model?
Haiku 5.5 is worth trying for document work, data extraction, and computer use. Anthropic reports strong results on these tasks, but your own data may produce different results.
Luna is a good option for high-volume calls and long prompts. OpenAI also offers a Decisions API using Luna for fixed-choice classification and scoring.
Sol 6.1 costs twenty times Luna's short-prompt rate. It scored higher on terminal coding in the comparison, so it may be worth testing for difficult repository tasks.
The benchmark runs were not identical. Sol and Luna's Terminal-Bench scores came from a public leaderboard; Haiku's came from Anthropic's system card with a different agent and setup. Treat the numbers as a starting point, not a direct contest.
Check the cost on your prompts
For a request with 150K input and 5K output tokens, Kingy's calculation is:
- Haiku 5.5: $0.0875
- GPT-6 Luna: $0.0175
- GPT-6.1 Sol: $0.35
These estimates assume equal token use, no caching, and no retries or tool charges. System prompts and conversation history count toward the input, so check actual requests before estimating a bill.
Try 30 to 50 real tasks with the same prompts and tools. Check extracted facts, run coding tests, and verify the final state of computer-use tasks. Compare cost per successful result, including retries and time spent correcting the output.
I'd start with Luna for long or frequent requests, Haiku for document and desktop tasks, and Sol for hard coding tasks if the extra cost is worth it. Keep your current model as a baseline.
Sources: Kingy's comparison, Anthropic Haiku 5.5, Haiku system card, OpenAI Luna specs, OpenAI Sol 6.1 specs, OpenAI pricing.