Claude Haiku 5.5 on Vidman AI
claude-haiku-5-5
Pricing
₹ per 1M tokens · as of 16 October 2026| Input | ₹9.60 |
| Output | ₹48 |
| Cached input | ₹0.96 |
| Cache write · 5 min | ₹12 |
| Cache write · 1 hour | ₹19 |
Flat rate: every request bills the same. Rupee prices convert at the platform exchange rate billing uses (₹96/USD today). Full catalog and comparisons on the pricing comparison page.
About this model
Anthropic's fastest, lowest-cost Claude 5.5 model for high-volume, latency-sensitive work: classification, routing, extraction, summaries and subagent tasks. Adaptive thinking with adjustable effort.
Not sure? Vidman AI Adaptive routes to the best model per request, including this one.
Model specifications
API data- Modality
- Text
- Context
- 10,00,000 tokens
- Maximum output
- 1,28,000 tokens
- Reasoning
- Optional
- Effort levels
- low, medium, high, xhigh, max
- Default effort
- medium
- Caching
- Platform managed
- Cache minimum
- 512 tokens
- Cache TTL
- 5m / 1h
- Route
- Vidman AI managed
From the serverless catalog, fetched 16 October 2026.
Measured performance
API observations over 7d; not an SLA.
- TTFT p50
- 6,000 ms
- TTFT p95
- 12,000 ms
- Throughput p50
- 500 tok/s
- Throughput average
- 286.8 tok/s
curl https://api.vidman.ai/v1/chat/completions \
-H "x-api-key: $VIDMAN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-haiku-5-5",
"messages": [{ "role": "user", "content": "Hello" }]
}'Existing OpenAI SDK code works by changing the base URL to https://api.vidman.ai/v1 and the key.
Model card
Claude Haiku 5.5 is the smallest and fastest model in Anthropic's Claude 5.5 family, built for high-volume, latency-sensitive work: classification, routing, extraction, summaries, live customer support and subagent tasks alongside Claude Opus 5.5 and Claude Sonnet 5.5. It has a 1M-token context window and returns up to 128K output tokens. Anthropic describes it as the cheapest, fastest and most capable small model it has released, costing around 75% less to run than Claude Haiku 4.5 on average.
Capabilities
- Adaptive thinking, on by default, with an effort setting: the first Haiku model with one. At lower effort it thinks less and can skip thinking entirely on simple requests.
- Reasoning off:
reasoning_effort: "none"(orthinking: {"type": "disabled"}on the Messages API) turns thinking off for the fastest, cheapest answers. - Long context: a 1M-token context window and up to 128K output tokens per request.
- Multimodal input: text and image input, with text output. Images up to 8,000 px on the longest edge; above 20 images in one request, each image may be at most 2,000 px.
- Tool use:
tool_choiceofauto,none,requiredand a named tool all work, and it calls several tools in parallel. A forced tool call starts with the call itself and no thinking block. - Structured output:
response_formatwith a JSON schema returns schema-valid JSON. - Prompt caching: caching engages from a 512-token cacheable prefix, and cache reads cost a tenth of the input price.
Pricing by prompt length
Haiku 5.5 is priced by the length of each request's prompt, counting input, cache reads and cache writes together. Prompts up to 100,000 tokens pay the listed prices. A request whose prompt is over 100,000 tokens pays five times the listed price on every side, output included, for the whole request. Anthropic notes that prompts up to 100,000 tokens make up around 90% of requests to the previous Haiku model.
Benchmark results
Results reported in Anthropic's launch materials and the Claude Haiku 5.5 System Card.
| Benchmark | Claude Haiku 5.5 | Claude Haiku 4.5 | Claude Sonnet 5.5 (reference) |
|---|---|---|---|
| GDPval-AA v2.1 (knowledge work) | 1620 Elo | 735 Elo | 1840 Elo |
| AA-Briefcase v1.1 (long-horizon knowledge work) | 1578 Elo | 614 Elo | 1824 Elo |
| OSWorld 2.1 (computer use, offline subset) | 72.4% | 15.7% | 83.9% |
| Humanity's Last Exam (no tools) | 45.9% | 10.2% | 56.9% |
| Humanity's Last Exam (with tools) | 57.4% | 18.7% | 64.5% |
| Terminal-Bench 4.0 (agentic coding) | 39.2% | 0.0% | 70.6% |
| FrontierCode 1.1 Main (agentic coding) | 46.4% | : | 52.1% (xhigh) |
| Chartography (chart recognition, no tools) | 46.4% | 6.4% | 61.6% |
Working with the model
Early testers highlighted speed and accuracy on short, repeated work:
- Asana measured over 30% lower latency for task completions and up to 2.5x faster inference per agent turn than the model it uses today.
- HubSpot reported the best score it has seen on its CRM evaluation suite, 92.8% averaged over three runs, with the fastest completion and the lowest false-positive rate on a CRM audit task.
- Box saw Haiku 5.5 score 11 points higher than Haiku 4.5 at about half the latency.
Behaviours to plan for when moving from Claude Haiku 4.5:
- Recount tokens. It uses the newer tokenizer, so the same text counts as roughly 30% more tokens, and large images cost more visual tokens.
- Thinking counts toward
max_tokens. A small limit can stop after a thinking block and before any text; raise it or lower the effort. - No assistant prefill. End the conversation with a user turn.
- Sampling parameters are ignored.
temperature,top_pandtop_kare not applied on this model; guide it with the prompt instead. - Thinking blocks are tied to the conversation. Keep conversation history append-only when replaying Haiku 5.5 thinking blocks.
Safety and evaluations
Anthropic's pre-deployment evaluation is documented in the Claude Haiku 5.5 System Card. It shows major improvements over Haiku 4.5 across almost all of Anthropic's alignment evaluations, with far fewer instances of misaligned behaviour and a lower willingness to cooperate with misuse. Its cybersecurity safeguards are more restrictive than Haiku 4.5's but less restrictive than those on other recent models, and its biology safeguards match Sonnet 5, Sonnet 5.5 and Opus 5. These target a narrow set of high-risk requests; some requests in those areas may be declined.