Claude 3.5 HaikuDeprecated
Benchmarks
independent evaluations · source: Artificial Analysis| Benchmark | Best score | Effort | Cost / task | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CritPt | 0% | default | — | |||||||||||||||||
| GDPval v2Frozen Aug 18, 2026 | 457.0 | default | — | |||||||||||||||||
| ||||||||||||||||||||
| GPQA Diamond | 41% | default | — | |||||||||||||||||
| ||||||||||||||||||||
| Humanity's Last Exam | 4% | default | — | |||||||||||||||||
| IFBench | 43% | default | — | |||||||||||||||||
| LiveCodeBenchArtificial Analysis did not report this score in its latest run on Oct 14, 2026 | 31% | — | — | |||||||||||||||||
| Long-Context Reasoning | 27% | default | — | |||||||||||||||||
| MMLU-ProFrozen Aug 18, 2026 | 63% | default | — | |||||||||||||||||
| MMMU-Pro | 46% | default | — | |||||||||||||||||
| Omniscience | -22.5 | default | — | |||||||||||||||||
| Omniscience Accuracy | 0.1 | default | — | |||||||||||||||||
| Omniscience Non Hallucination | 0.6 | default | — | |||||||||||||||||
| SciCodeArtificial Analysis did not report this score in its latest run on Oct 14, 2026 | 27% | — | — | |||||||||||||||||
| TerminalBench Hard | 2% | default | — | |||||||||||||||||
| TerminalBench v2.1 | 10% | default | — | |||||||||||||||||
| ||||||||||||||||||||
| τ-bench BankingArtificial Analysis did not report this score in its latest run on Oct 14, 2026 | 6% | — | — | |||||||||||||||||
| τ²-bench | 25% | default | — | |||||||||||||||||
Pricing by serving provider
standard tier · per 1M tokens · click a provider to expandInput pricing runs from $0.80 to $1.00 per 1M tokens across 3 providers, so the dearest route costs 25% more than the cheapest for the same model.
| Serving provider | Input /1M | Output /1M | Endpoints | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| VercelRouted quoteVercel reseller quote | $0.80 | $4.00 | 1 | ||||||||||||||||
Standardvercel_ai_gateway/anthropic/claude-3.5-haiku
| |||||||||||||||||||
| Replicate | $1.00 | $5.00 | 1 | ||||||||||||||||
Standardreplicate/anthropic/claude-3.5-haiku
| |||||||||||||||||||
| Vertex AI Anthropic Models | $1.00 | $5.00 | 1 | ||||||||||||||||
Standardvertex_ai/claude-3-5-haiku
| |||||||||||||||||||
Price history
listed seller prices, input and output per 1M tokensCost calculator
estimate your monthly spend on this modelModel IDs
copy the exact identifier for your platformFrequently asked questions
Claude 3.5 Haiku pricing, context and availabilityIs Claude 3.5 Haiku still available?
Claude 3.5 Haiku was deprecated on 2026-07-05, but 3 serving providers still list it.
How much does Claude 3.5 Haiku cost?
Claude 3.5 Haiku costs $0.80 per 1M input tokens as a Vercel reseller quote, and $4.00 per 1M output tokens as a Vercel reseller quote. A Vercel reseller quote is the price Vercel lists for serving this model, not the maker list price, so it moves when Vercel re-routes. Across 3 serving providers, input prices range from $0.80 to $1.00 per 1M tokens.
What is the context window of Claude 3.5 Haiku?
Claude 3.5 Haiku accepts up to 200K tokens of context and can return up to 8K output tokens.
Which providers serve Claude 3.5 Haiku?
Claude 3.5 Haiku is available from 3 serving providers, each with its own pricing and model ID. Vercel is currently the cheapest, as a routed quote.
What can Claude 3.5 Haiku do?
Claude 3.5 Haiku supports image input (vision), tool use, prompt caching and structured output.
Other Anthropic models
compare pricing across the Anthropic lineupPopular comparisons
head-to-head pages featuring Claude 3.5 Haiku
Claude 3.5 Haiku vs GPT-4oEvery weekday
AI moves fast. Here's your debrief.
News, analysis, tools, and more.
For people who build with AI