Claude Haiku 5.5 vs GPT-6 Luna

Claude Haiku 5.5: $0.10 input and $0.50 output per million tokens. GPT-6 Luna: $0.10 input and $0.50 output per million tokens. The cheaper choice depends on your input-to-output ratio; a model can have a lower input rate but a higher output rate.

For an example workload of 20 million input tokens and 5 million output tokens per month: Claude Haiku 5.5: $4.50. GPT-6 Luna: $4.50. These are token-only estimates using the displayed rates, without caching discounts, additional tool charges or taxes. Actual bills depend on usage and endpoint pricing.

Claude Haiku 5.5: 1M tokens of context; maximum output 128K tokens. GPT-6 Luna: 1.1M tokens of context; maximum output 128K tokens. Context capacity and maximum output are separate limits. A larger context window allows more material in a request but does not guarantee more accurate answers.

Claude Haiku 5.5: Artificial Analysis SciCode not listed. GPT-6 Luna: Artificial Analysis SciCode 54.6%. Compare coding results within the same evaluation, then test representative tasks from your own codebase. A missing score is not a zero, and a single benchmark does not establish the best model for every programming task.

These evaluations measure different tasks and use different scales. Arena Elo and DesignArena ratings must remain separate, as must Artificial Analysis Intelligence Index and MMLU-Pro. Compare models within the same evaluation rather than combining scores. Missing results are marked Not listed, not treated as measured zeroes.

Claude Haiku 5.5: Artificial Analysis output speed not listed; time to first token not listed. GPT-6 Luna: Artificial Analysis output speed 141 tokens/s; time to first token 0.1 seconds. Output speed measures generation throughput, while time to first token measures the initial wait. Real-world latency also depends on the API provider, request size and load; missing measurements cannot establish a speed winner.

Claude Haiku 5.5: listed input formats: text, image, file; output formats: text. GPT-6 Luna: listed input formats: file, image, text; output formats: text. Input and output support are different capabilities: accepting an image does not mean a model generates images. Confirm formats and file limits for the endpoint you plan to use.

Claude Haiku 5.5: listed features: Function calling, Structured outputs, JSON mode, Reasoning, Prompt caching. GPT-6 Luna: listed features: Function calling, Structured outputs, JSON mode, Reasoning, Deterministic seed, Prompt caching. Only recorded capabilities are shown; an unlisted feature is not proof that it is unsupported. Availability can vary by API provider, so confirm tool calling and JSON schema support for your chosen endpoint.

Claude Haiku 5.5: cached input price $0.01 per million tokens. GPT-6 Luna: cached input price $0.01 per million tokens. Caching may reduce charges for eligible repeated input when supported by the endpoint. A listed cached rate does not guarantee every request qualifies; cache duration, minimum token counts and any write charges depend on the provider.

Claude Haiku 5.5: catalog status active; API identifier claude-haiku-5.5; listed API hosts Google, Claude Platform on AWS, Azure, Anthropic, Amazon Bedrock. GPT-6 Luna: catalog status active; API identifier gpt-6-luna; listed API hosts OpenAI, Azure, Amazon Bedrock. Catalog status is not a live availability check. Before integrating, verify the provider's documentation, endpoint access, rate limits, regional availability and data-handling terms.