Today's price · per 1M tokens
Input
$0.100
per 1M tokens
Output
$0.500
per 1M tokens
Blended
$0.200
blended $/1M — 3:1 weighted input:output
Cached input
$0.010
10% of input — prompt caching
Price receipt
We read Anthropic's own pricing page on and found Claude Haiku 5.5 listed with a price on the same row — that read is where the number above comes from, recorded as a new model. We keep the page text we read; its content hash is 2ec216b67b.
Overview
Anthropic's newest Haiku, launched 2026-10-07 as the fastest and lowest-cost model of the Claude 5.5 family, built for high-volume work such as classification, summaries and coding sub-agents; it is the first Haiku with an adjustable effort setting. Priced by prompt length: prompts up to 100K tokens bill at $0.10/$0.50 per 1M tokens (cache reads $0.01); prompts above 100K tokens bill at the long-context tier of $0.50/$2.50 (cache reads $0.05). Batch API halves both tiers. 1M-token context.
Capabilities
struck through = not supportedBenchmark performance
accuracy % · higher is betterSources: Epoch AI (epoch.ai/benchmarks, CC BY 4.0) (GPQA Diamond) · Artificial Analysis (text-only, no tools, max effort) (Humanity's Last Exam)
Source: Epoch AI (epoch.ai/benchmarks), CC BY 4.0
- AA Intelligence Index: 36.5 (max effort) (Artificial Analysis composite across reasoning, knowledge and coding evals)
- AA-Omniscience Index: 12.3 (max effort) (−100–100)
- GDPval-AA v2: 1629 ELO (Real-world work tasks, human baseline = 1000)
- Arena (arena.ai): 1420.8 ELO (Human preference)
- Arena WebDev: 1595.4 ELO (Human preference on building web apps)
- Arena Agent: -0.006 (Score from real agent sessions (IPS); higher is better)
- Arena Agent cost per task: $0.09 (Median measured cost of one real Agent Arena task)
- Epoch Capabilities Index: 160.3 (CI 158.2–162.7) (Epoch AI's index fitted across many benchmarks)
- FrontierMath Tiers 1–3: 75.1% (max effort) (Research-level maths (funded by OpenAI, run by Epoch); best run across the effort settings Epoch tested)
- FrontierMath Tier 4: 46.3% (max effort) (Hardest research maths; best run across the effort settings Epoch tested)
Source: Artificial Analysis (artificialanalysis.ai)
- HLE-Diamond: 29.6% (max effort, measured 2026-11-22) — A revised 1,000-question set of Humanity's Last Exam, half reasoning, half knowledge
- CritPt v2, hard subset: 47.5% (xhigh effort, measured 2026-11-22) — The hardest research-level physics problems in CritPt v2, solved with a code sandbox
- FrontierCode-AA 2.0: 34.9% (max effort, measured 2026-11-22) — Coding tasks in real open-source repositories, graded on whether a maintainer would merge the change
- Terminal-Bench-Science: 20% (max effort, measured 2026-11-22) — Expert-written science research workflows completed in a terminal
- GDP.pdf: 20.8% (max effort, measured 2026-11-22) — Tasks over long professional documents; share of tasks that pass every criterion
- AutomationBench-AA: 36% (xhigh effort, measured 2026-11-22) — Multi-step work across simulated business apps, without breaking the rules set
- MLCR-AA: 35% (max effort, measured 2026-11-22) — Reasoning over long, fragmented medical records; pass rate
- AA-LCR (long-context reasoning): 82.7% (max effort, measured 2026-11-22)
- Terminal-Bench 4.0: 32.8% (max effort, measured 2026-11-22)
- SciCode: 55% (max effort, measured 2026-11-22)
- Cost to run the AA Intelligence Index: $0.411 per task, 14.7 min per task (max effort)
- Measured by AA: 257.1 tokens/sec output, 170.15s to first token (max effort)
- Ranks #59 of 207 comparably-measured models by percentile score, across 6 independent measurements
- Ranks #6 of 207 comparably-tested models by normalized performance per dollar
- Strongest at Humanity's Last Exam — 44.4%, #33 of 81
257.1 tokens/sec output 170.15s latency to first token (TTFT)
Throughput: Artificial Analysis, measured 2026-11-22 · Source: Artificial Analysis (artificialanalysis.ai)
Source: Artificial Analysis, Artificial Analysis (text-only, no tools, max effort) · updated · See full rankings →
Specifications
- Provider
- Anthropic
- Context window
- 1M tokens
- Modality
- text, image
- Parameters
- Proprietary
- Open source
- No — proprietary
- Released
- Status
- Current
- Last updated
- Tags
Availability verified: — listed on Anthropic's own page