Breaking

Anthropic Launches Claude Haiku 5.5 to Challenge OpenAI's GPT-6 Luna on Low-Cost AI Agents

October 7, 2026 at 23:16 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic has released Claude Haiku 5.5, a small, low-cost model aimed at high-volume work, setting up a direct comparison with OpenAI's GPT-6 Luna. The contest matters because both models begin at the same API price for short prompts, yet they diverge on benchmarks, long-context economics and the developer ecosystems around agents.

Oct. 7, 2026 · Small-model race

Claude Haiku 5.5 vs GPT-6 Luna

Anthropic's new small model enters the same entry-price band as GPT-6 Luna, but the real contest is benchmark fit, long-context pricing and agent workflow economics.

The opening price looks identical

$0.10

Input per 1M tokens at the entry tier

$0.50

Output per 1M tokens at the entry tier

1M / 1.05M

Context windows for Haiku 5.5 and Luna

100K / 272K

Prompt thresholds before higher long-context rates

Terminal-Bench 4.0 comparison

Same unit, drawn as proportional stacked columns.

39.2%

Claude Haiku 5.5

16.4%

GPT-6 Luna

Best fit

Short, repeated work: routing, extraction, summarization, compaction, browser use and subagents.

Watch point

Long single prompts change the bill: Haiku steps up after 100K tokens, while Luna steps up after 272K input tokens.

The buying decision shifts to workload shape

The same sticker price now hides different economics. Teams running agent loops will need to test their real prompts, cache behavior and tool stack before switching.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.