Physix Frontier · News Briefing Card (QbitAI · Oct 8, 2026)

Anthropic launches Claude Haiku 5.5, beating GPT-6 Luna on benchmarks

KEY FACTS

Anthropic released Claude Haiku 5.5, which outperforms GPT-6 Luna on every official benchmark.
Haiku 5.5 is priced to match GPT-6 Luna, with input and output prices dropping 90% within 100,000 tokens.
Haiku 5.5 is the first Haiku model to support effort adjustment, offering five configuration levels.
Haiku 5.5 switched tokenizers, increasing token consumption for the same tasks, so actual savings fall short of the list-price discount.
On Terminal-Bench 4.0, Haiku 5.5 scored 39.2%, while Sonnet 5.5 scored 70.6%.

KEY DATA

72.4%OSWorld 2.1 Max level accuracy

$0.61OSWorld 2.1 Max level cost per run

39.2%Terminal-Bench 4.0 score

$100Max 5x monthly API credit

PHYSIX OBSERVATION

Haiku 5.5 pushes the price war down to the small-model tier, but the tokenizer switch shrinks its real cost advantage, and users still need to change five interfaces to migrate. It suits the execution layer rather than replacing Sonnet; Anthropic is subsidizing API experiments with subscription credits, aiming to lock in the developer ecosystem.

Source: QbitAI report