Claude Haiku 5.5: Anthropic's small model gets a 75% price cut and an effort dial
$0.10/$0.50 per million input/output tokens, an adjustable effort setting for the first time in the Haiku line, and customer-reported scores that beat Haiku 4.5 by double digits on internal evals -- aimed squarely at high-volume, cost-sensitive work.
Anthropic released Claude Haiku 5.5 on October 7, calling it "the cheapest, fastest, and most capable small model we've ever released." Pricing drops to $0.10/$0.50 per million input/output tokens for prompts under 100k tokens ($0.50/$2.50 above that threshold) -- Anthropic says that works out to roughly 75% less than Haiku 4.5 on average workloads. It's the first Haiku-class model to ship with an adjustable effort setting, previously reserved for Anthropic's larger models.
What customers are reporting
Three companies published specific before/after numbers in Anthropic's own release: HubSpot said Haiku 5.5 "got the best score we've seen on this suite yet, at 92.8% averaged over three runs"; Box reported it "scored 11 points higher than Haiku 4.5 at about half the latency" in early testing; AlphaSense called the gain "a statistically significant improvement over Haiku 4.5: 0.84 vs. 0.76." Asana reported a 30%-plus latency reduction on task completions versus its prior model. These are customer-reported, not independently audited, but they're specific numbers tied to named companies rather than vague praise.
The number that matters for a spend decision
On Anthropic's own benchmark table, Haiku 5.5 jumps from Haiku 4.5's 0% to 39.2% on Terminal-Bench 4.0, and from 15.7% to 72.4% on the offline OSWorld 2.1 subset -- both agentic-task benchmarks, not knowledge tests, which lines up with Anthropic positioning this model as a subagent alongside Opus 5.5 and Sonnet 5.5 rather than a standalone chat model. For a team already routing high-volume, low-complexity work to Haiku 4.5, the actual question isn't the benchmark delta -- it's whether that 75% price cut holds on your own traffic mix once you swap the model ID, since list-price cuts and benchmark gains on Anthropic's chosen tasks don't automatically transfer to a different workload.