Claude Haiku 5.5 is Anthropic’s small-model edition for high-volume, latency-sensitive tasks. Its main uses include classification, routing, information extraction, summarisation, conversation compaction and focused subagent work. It accepts text and images and generates text. A useful brief gives it a narrow question and an explicit result: a category, a defined set of fields, a short summary or the relevant evidence for another part of a larger task. Keep the source identifiers with the result so an application can connect the extracted information to its original material.
Anthropic positions this edition for live customer support, browser use and short coding subtasks as well as repetitive analysis. It can complement Claude Sonnet 5.5 or Claude Opus 5.5 in a workflow where a larger model handles an ambitious project and Haiku handles a bounded lookup or summary. For a support task, supply the question and the permitted reference material, then specify the response tone and what must be returned. For a coding subtask, identify the relevant files or excerpts and the exact question, rather than assigning an open-ended project without a completion condition.
The documented context window is one million tokens, with up to 128,000 output tokens in ordinary generation. Thinking tokens count towards the generation allowance, so the result needs room for both reasoning and visible text. Its reliable knowledge cutoff is June 2026; later information needs supplied evidence or a search tool. Anthropic separately documents a batch-only beta that can raise output to 300,000 tokens for this edition. That larger allowance requires the producer’s extended-output feature and is distinct from the ordinary limit. An asynchronous batch handles requests independently, making it useful for collections of documents or records whose results are needed after processing rather than immediately.
Adaptive thinking is enabled by default, and effort is the main control for reasoning depth. The levels are low, medium, high, xhigh and max, with medium as the default. Anthropic recommends medium for most work, high for more sustained tasks or strict instruction following, and low for simpler, short requests. Thinking can be disabled at low, medium or high; disabling it at xhigh or max is rejected. A prompt asking for a direct answer does not substitute for that control. Manual budget_tokens settings from Haiku 4.5 do not transfer to this edition, and temperature, top_p and top_k should be omitted in the documented interface.
Haiku 5.5 uses the newer tokenizer introduced in earlier Claude generations, and Anthropic estimates that the same text can use roughly 30% more tokens than on Haiku 4.5, depending on its content. It also retains thinking blocks from earlier assistant turns, so conversation size can grow beyond the tokenizer change alone. Its minimum cacheable prompt length is 512 tokens. Keep earlier turns unchanged when retaining reasoning, and put new instructions in the newest turn. Changing top-level effort invalidates the conversation’s message cache; the producer documents a separate beta for changing effort within adaptive conversations. None of these capabilities establishes a cache hit for a particular request.
For retrieval, the model’s prompting guide recommends supplying the current date with the search tool and keeping the original system and tool definitions stable. Give a fresh date in the newest turn if the conversation continues on a later day. State what must be found and preserve the returned sources before asking for a conclusion. For example, an illustrative task could ask for the documented behaviour of a specified software version, followed by a short explanation grounded in the retrieved passage. The application supplies the search tool and returns its actual results; the model’s proposed lookup is not evidence that the lookup happened.
When structured JSON output is combined with tools, the guide documents a specific interaction: with thinking disabled, Haiku can skip a necessary tool call. Anthropic recommends adaptive thinking for such requests, or separating the final JSON requirement from the tool step. Define the output fields and make clear that the final answer follows the tool results. A data-extraction brief can name the required fields and how to represent missing values, while a lookup brief can first request the evidence needed to fill them. Read response blocks by their type, since the answer need not be the first block and thinking text is omitted by default.
For longer agent prompts, Anthropic notes that low effort can stop early, skip a search or omit a check. Its guide recommends explicit completion instructions and real checks for code changes before they are reported as finished. Prepare the expected result and the evidence that would demonstrate it: the selected passages for a report, the changed behaviour for a code proposal or the fields extracted from a document. The guide also distinguishes a user’s mid-task message from tool output: new user instructions belong in a user turn, while the returned tool evidence remains separate.
This edition has its own safeguards and can return a refusal for categories including cyber harm, biological harm, competing-model development or other prohibited uses. The producer does not document server-side fallback for Haiku 5.5. Its guide also notes that xhigh effort can occasionally end a multi-turn response without visible answer text, and that reasoning-like text can appear more often at low effort or with thinking disabled. These are specific behaviours to account for when a workflow requires a visible, structured result. For ordinary extraction or summary work, use a focused source, a defined output and the effort level appropriate to the task.
", "aboutText": "Claude Haiku 5.5 is Anthropic’s small-model edition for high-volume, latency-sensitive tasks. Its main uses include classification, routing, information extraction, summarisation, conversation compaction and focused subagent work. It accepts text and images and generates text. A useful brief gives it a narrow question and an explicit result: a category, a defined set of fields, a short summary or the relevant evidence for another part of a larger task. Keep the source identifiers with the result so an application can connect the extracted information to its original material.\n\nAnthropic positions this edition for live customer support, browser use and short coding subtasks as well as repetitive analysis. It can complement Claude Sonnet 5.5 or Claude Opus 5.5 in a workflow where a larger model handles an ambitious project and Haiku handles a bounded lookup or summary. For a support task, supply the question and the permitted reference material, then specify the response tone and what must be returned. For a coding subtask, identify the relevant files or excerpts and the exact question, rather than assigning an open-ended project without a completion condition.\n\nThe documented context window is one million tokens, with up to 128,000 output tokens in ordinary generation. Thinking tokens count towards the generation allowance, so the result needs room for both reasoning and visible text. Its reliable knowledge cutoff is June 2026; later information needs supplied evidence or a search tool. Anthropic separately documents a batch-only beta that can raise output to 300,000 tokens for this edition. That larger allowance requires the producer’s extended-output feature and is distinct from the ordinary limit. An asynchronous batch handles requests independently, making it useful for collections of documents or records whose results are needed after processing rather than immediately.\n\nAdaptive thinking is enabled by default, and effort is the main control for reasoning depth. The levels are low, medium, high, xhigh and max, with medium as the default. Anthropic recommends medium for most work, high for more sustained tasks or strict instruction following, and low for simpler, short requests. Thinking can be disabled at low, medium or high; disabling it at xhigh or max is rejected. A prompt asking for a direct answer does not substitute for that control. Manual budget_tokens settings from Haiku 4.5 do not transfer to this edition, and temperature, top_p and top_k should be omitted in the documented interface.\n\nHaiku 5.5 uses the newer tokenizer introduced in earlier Claude generations, and Anthropic estimates that the same text can use roughly 30% more tokens than on Haiku 4.5, depending on its content. It also retains thinking blocks from earlier assistant turns, so conversation size can grow beyond the tokenizer change alone. Its minimum cacheable prompt length is 512 tokens. Keep earlier turns unchanged when retaining reasoning, and put new instructions in the newest turn. Changing top-level effort invalidates the conversation’s message cache; the producer documents a separate beta for changing effort within adaptive conversations. None of these capabilities establishes a cache hit for a particular request.\n\nFor retrieval, the model’s prompting guide recommends supplying the current date with the search tool and keeping the original system and tool definitions stable. Give a fresh date in the newest turn if the conversation continues on a later day. State what must be found and preserve the returned sources before asking for a conclusion. For example, an illustrative task could ask for the documented behaviour of a specified software version, followed by a short explanation grounded in the retrieved passage. The application supplies the search tool and returns its actual results; the model’s proposed lookup is not evidence that the lookup happened.\n\nWhen structured JSON output is combined with tools, the guide documents a specific interaction: with thinking disabled, Haiku can skip a necessary tool call. Anthropic recommends adaptive thinking for such requests, or separating the final JSON requirement from the tool step. Define the output fields and make clear that the final answer follows the tool results. A data-extraction brief can name the required fields and how to represent missing values, while a lookup brief can first request the evidence needed to fill them. Read response blocks by their type, since the answer need not be the first block and thinking text is omitted by default.\n\nFor longer agent prompts, Anthropic notes that low effort can stop early, skip a search or omit a check. Its guide recommends explicit completion instructions and real checks for code changes before they are reported as finished. Prepare the expected result and the evidence that would demonstrate it: the selected passages for a report, the changed behaviour for a code proposal or the fields extracted from a document. The guide also distinguishes a user’s mid-task message from tool output: new user instructions belong in a user turn, while the returned tool evidence remains separate.\n\nThis edition has its own safeguards and can return a refusal for categories including cyber harm, biological harm, competing-model development or other prohibited uses. The producer does not document server-side fallback for Haiku 5.5. Its guide also notes that xhigh effort can occasionally end a multi-turn response without visible answer text, and that reasoning-like text can appear more often at low effort or with thinking disabled. These are specific behaviours to account for when a workflow requires a visible, structured result. For ordinary extraction or summary work, use a focused source, a defined output and the effort level appropriate to the task.", "editorialOverlay": { "copy_fields": null, "copy_sha256": null, "copy_source": null, "html_sha256": "6f565a1e69cf47516c7c710c25acd3d2c1d88145ceacdc06e85658f50e820104", "module_sha256": "ee9dc1c50b1e6a8fb7fa062030fcb4646b9aaaeb08be513b723a5a92a7290f06", "raw_model_unchanged": true, "schema": "ailoot-native-model-editorial-v1" }, "locale": "en", "methods": [ { "html": "anthropic/claude-haiku-5.5:batchClaude Haiku 5.5 is Anthropic’s small model for classification, extraction, summaries, routing and focused agent tasks. It combines text and image understanding with adaptive reasoning and is shaped for repetitive work that needs a prompt response.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
Claude Haiku 5.5 is Anthropic’s small-model edition for high-volume, latency-sensitive tasks. Its main uses include classification, routing, information extraction, summarisation, conversation compaction and focused subagent work. It accepts text and images and generates text. A useful brief gives it a narrow question and an explicit result: a category, a defined set of fields, a short summary or the relevant evidence for another part of a larger task. Keep the source identifiers with the result so an application can connect the extracted information to its original material.
Anthropic positions this edition for live customer support, browser use and short coding subtasks as well as repetitive analysis. It can complement Claude Sonnet 5.5 or Claude Opus 5.5 in a workflow where a larger model handles an ambitious project and Haiku handles a bounded lookup or summary. For a support task, supply the question and the permitted reference material, then specify the response tone and what must be returned. For a coding subtask, identify the relevant files or excerpts and the exact question, rather than assigning an open-ended project without a completion condition.
The documented context window is one million tokens, with up to 128,000 output tokens in ordinary generation. Thinking tokens count towards the generation allowance, so the result needs room for both reasoning and visible text. Its reliable knowledge cutoff is June 2026; later information needs supplied evidence or a search tool. Anthropic separately documents a batch-only beta that can raise output to 300,000 tokens for this edition. That larger allowance requires the producer’s extended-output feature and is distinct from the ordinary limit. An asynchronous batch handles requests independently, making it useful for collections of documents or records whose results are needed after processing rather than immediately.
Adaptive thinking is enabled by default, and effort is the main control for reasoning depth. The levels are low, medium, high, xhigh and max, with medium as the default. Anthropic recommends medium for most work, high for more sustained tasks or strict instruction following, and low for simpler, short requests. Thinking can be disabled at low, medium or high; disabling it at xhigh or max is rejected. A prompt asking for a direct answer does not substitute for that control. Manual budget_tokens settings from Haiku 4.5 do not transfer to this edition, and temperature, top_p and top_k should be omitted in the documented interface.
Haiku 5.5 uses the newer tokenizer introduced in earlier Claude generations, and Anthropic estimates that the same text can use roughly 30% more tokens than on Haiku 4.5, depending on its content. It also retains thinking blocks from earlier assistant turns, so conversation size can grow beyond the tokenizer change alone. Its minimum cacheable prompt length is 512 tokens. Keep earlier turns unchanged when retaining reasoning, and put new instructions in the newest turn. Changing top-level effort invalidates the conversation’s message cache; the producer documents a separate beta for changing effort within adaptive conversations. None of these capabilities establishes a cache hit for a particular request.
For retrieval, the model’s prompting guide recommends supplying the current date with the search tool and keeping the original system and tool definitions stable. Give a fresh date in the newest turn if the conversation continues on a later day. State what must be found and preserve the returned sources before asking for a conclusion. For example, an illustrative task could ask for the documented behaviour of a specified software version, followed by a short explanation grounded in the retrieved passage. The application supplies the search tool and returns its actual results; the model’s proposed lookup is not evidence that the lookup happened.
When structured JSON output is combined with tools, the guide documents a specific interaction: with thinking disabled, Haiku can skip a necessary tool call. Anthropic recommends adaptive thinking for such requests, or separating the final JSON requirement from the tool step. Define the output fields and make clear that the final answer follows the tool results. A data-extraction brief can name the required fields and how to represent missing values, while a lookup brief can first request the evidence needed to fill them. Read response blocks by their type, since the answer need not be the first block and thinking text is omitted by default.
For longer agent prompts, Anthropic notes that low effort can stop early, skip a search or omit a check. Its guide recommends explicit completion instructions and real checks for code changes before they are reported as finished. Prepare the expected result and the evidence that would demonstrate it: the selected passages for a report, the changed behaviour for a code proposal or the fields extracted from a document. The guide also distinguishes a user’s mid-task message from tool output: new user instructions belong in a user turn, while the returned tool evidence remains separate.
This edition has its own safeguards and can return a refusal for categories including cyber harm, biological harm, competing-model development or other prohibited uses. The producer does not document server-side fallback for Haiku 5.5. Its guide also notes that xhigh effort can occasionally end a multi-turn response without visible answer text, and that reasoning-like text can appear more often at low effort or with thinking disabled. These are specific behaviours to account for when a workflow requires a visible, structured result. For ordinary extraction or summary work, use a focused source, a defined output and the effort level appropriate to the task.
POST /v1/batchestexttext| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldendpoint | Typestring | RequirementRequired | Values & limits /v1/chat/completions | Default— |
Fieldmodel | Typestring | RequirementRequired | Values & limits anthropic/claude-haiku-5.5:batch | Default— |
Fieldrequests | Typebatch_item[] | RequirementRequired | Values & limits Minimum items: 1 | Default— |
Fieldrequests[].custom_id | Typestring | RequirementRequired | Values & limits | Default— |
Fieldrequests[].body.messages | Typemessage[] | RequirementRequired | Values & limits | Default— |
Fieldcompletion_window | Typestring | RequirementOptional | Values & limits 24h | Default— |
Fieldprovider | Typeobject|null | RequirementOptional | Values & limits | Default— |
Fieldrequests[].body.model | Typestring | RequirementOptional | Values & limits anthropic/claude-haiku-5.5:batch, anthropic/claude-haiku-5.5 | Default— |
Fieldprovider.only | Typestring[] | RequirementOptional | Values & limits anthropic | Default— |
Fieldprovider.allow_fallbacks | Typeboolean|null | RequirementOptional | Values & limits false, null | Default— |
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractendpoint, model, requests |
Parametermaterial_types | Contracttext |
Parameterrepresentation | Contractinline requests[]; each item has custom_id and body |
Parameterrequest_endpoint_profile | Contract/v1/chat/completions |
Parameterfile_material | ContractCaller may read actual UTF-8 JSONL locally into requests[]. No upload purpose, input_file_id or file reference is used. Missing/empty customer material stops preparation before POST. |
Parameterserialization_order | Contractendpoint, model, provider (if supplied), completion_window (if supplied), requests last |
Parameteroptional_controls | Contractprovider.only, provider.allow_fallbacks, completion_window |
Parameterschema_scope | ContractText Chat Completions bodies in the documented inline Batch envelope. |
Lifecycle: asynchronous
GET /v1/batches/{operation_id} · actual original create response.id; same user/key/group
Tariff references do not add request options. Supported parameters are listed for the selected method.
Match each charge to the conditions shown. Use one rate for each set of alternative configurations.
| Configuration | Rate schedule | Context tokens, more than | Context tokens, up to and including | Charge |
|---|---|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Context tokens, more than— | Context tokens, up to and includingNo upper bound | Charge Input · Input tokens $0.0001 / 1K tokens Output · Output tokens $0.0005 / 1K tokens Input · Cache read $0.00001 / 1K tokens Input · Cache write $0.000125 / 1K tokens Input · Cache write, 1 hour $0.0002 / 1K tokens |
| ConfigurationRequest-length tier 1 | Rate schedule— | Context tokens, more than0 | Context tokens, up to and including100,000 | Charge Input · Input tokens $0.0001 / 1K tokens Output · Output tokens $0.0005 / 1K tokens Input · Cache read $0.00001 / 1K tokens Input · Cache write $0.000125 / 1K tokens Input · Cache write, 1 hour $0.0002 / 1K tokens |
| ConfigurationRequest-length tier 2 | Rate schedule— | Context tokens, more than100,000 | Context tokens, up to and includingNo upper bound | Charge Input · Input tokens $0.0005 / 1K tokens Output · Output tokens $0.0025 / 1K tokens Input · Cache read $0.00005 / 1K tokens Input · Cache write $0.000625 / 1K tokens Input · Cache write, 1 hour $0.001 / 1K tokens |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
set -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\npython3 - <<'PY_REQUEST'\nimport json\nfrom pathlib import Path\nbody=json.loads('{"endpoint":"/v1/chat/completions","model":"anthropic/claude-haiku-5.5:batch","requests":[{"custom_id":"req-0001","body":{"messages":[{"role":"user","content":"Hello!"}],"max_tokens":16}}]}')\nPath("batch-request.json").write_text(json.dumps(body,ensure_ascii=False),encoding="utf-8")\nPY_REQUEST\ncurl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/batches \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -H 'Content-Type: application/json' \\\n --data-binary @batch-request.json -D batch-response-headers.txt -o batch-create-response.json\nimport json, os\nfrom pathlib import Path\nfrom urllib.request import Request, urlopen\n\ndef prepare_requests(customer_path):\n path = Path(customer_path)\n if not path.is_file(): raise ValueError("Customer JSONL file is missing")\n items = [json.loads(line) for line in path.read_text(encoding="utf-8").splitlines() if line.strip()]\n if not items: raise ValueError("Customer JSONL contains no requests")\n ids = set()\n for item in items:\n if set(item) != {"custom_id", "body"}: raise ValueError("Each item needs custom_id and body")\n cid = item["custom_id"]\n if not isinstance(cid, str) or not cid.strip() or cid in ids: raise ValueError("custom_id must be nonempty and unique")\n ids.add(cid)\n if not isinstance(item["body"], dict) or not item["body"].get("messages"): raise ValueError("Supply actual text messages")\n supplied_model = item["body"].get("model")\n if supplied_model is not None and supplied_model not in ['anthropic/claude-haiku-5.5:batch', 'anthropic/claude-haiku-5.5']: raise ValueError("body.model must be an exact documented caller or wire ID")\n return {"endpoint":"/v1/chat/completions", "model":"anthropic/claude-haiku-5.5:batch", "requests":items}\n\nif __name__ == "__main__":\n material = os.environ.get("AILOOT_BATCH_REQUESTS_JSONL", "")\n if not material: raise SystemExit("Set AILOOT_BATCH_REQUESTS_JSONL to your actual customer file")\n body = prepare_requests(material)\n key = os.environ.get("AILOOT_API_KEY", "")\n if not key: raise SystemExit("Set AILOOT_API_KEY")\n request = Request("https://api.ailoot.ai/v1/batches", data=json.dumps(body, ensure_ascii=False).encode("utf-8"), headers={"Authorization":"Bearer " + key, "Content-Type":"application/json"}, method="POST")\n with urlopen(request) as response:\n raw = response.read()\n Path("batch-create-response.json").write_bytes(raw)\n print(raw.decode("utf-8"))\nimport json\nfrom pathlib import Path\nraw = Path("batch-status-response.json").read_bytes()\nresponse = json.loads(raw)\nif response.get("status") != "completed":\n raise SystemExit("Keep the original operation; read its status without resubmitting")\nresults = response.get("results")\nif not isinstance(results, list) or not results:\n raise SystemExit("Preserve the original response; completed results are missing")\nPath("batch-results.jsonl").write_text("".join(json.dumps(item, ensure_ascii=False) + "\\n" for item in results), encoding="utf-8")\n# Preserve batch-status-response.json with aggregate usage and every result/error.\ninline results[].response.body JSON
GET /v1/batches/{operation_id}set -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\nBATCH_OPERATION_ID="$(python3 - <<'PY_ID'\nimport json\nfrom pathlib import Path\nfrom urllib.parse import quote\nresponse = json.loads(Path("batch-create-response.json").read_text(encoding="utf-8"))\noperation = response.get("id")\nif not isinstance(operation, str) or not operation: raise SystemExit("No actual original operation ID; preserve the response and do not resubmit")\nprint(quote(operation, safe=""))\nPY_ID\n)"\ncurl --fail-with-body --silent --show-error "https://api.ailoot.ai/v1/batches/$BATCH_OPERATION_ID" \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -o batch-status-response.json\nKeep every complete returned response, file and content part.
Preserve the full original status JSON, every result/custom_id and complete response or error body. Save completed results[] as JSONL without filtering.
Preserve the original operation ID. 503 result_pending_check has X-Should-Retry:false; never resubmit an unknown original batch.
| Field | Type |
|---|---|
Fieldresults[].custom_id | Typestring |
Fieldresults[].response.body.choices[].message.content | Typestring|null |
Fieldresults[].error | Typeobject|null |
Fieldusage | Typeobject|null |
Fieldid | Typestring |
Fieldmodel | Typestring |
Fieldendpoint | Typestring |
Fieldstatus | Typestring |
Fieldrequest_counts | Typeobject |
Fieldresults[].id | Typestring |
Fieldresults[].response.status_code | Typeinteger |
Fieldresults[].response.request_id | Typestring |
| Response | Condition | Next step |
|---|---|---|
Response400 · invalid_request_error | ConditionInvalid documented inline envelope, duplicate keys, item identity, or configured model capabilities conflict. | Next stepCorrect the indicated fields before any new submission. |
Response401 · authentication_error | ConditionInvalid API key. | Next stepUse the authorized Ailoot key. |
Response403 · permission_error | ConditionBatch group unavailable. | Next stepCheck key/group entitlement. |
Response422 · invalid_request_error | ConditionBatch uses balance billing; subscription creation is rejected. | Next stepUse the authorized balance-billed group. |
Response422 · price_unavailable | ConditionBatch input cannot be quoted. | Next stepPreserve response; use documented family input. |
Response402 · billing_error | ConditionBatch reserve unavailable. | Next stepResolve admission before a new request. |
Response503 · price_unavailable | ConditionBatch method price/native rates unavailable. | Next stepPreserve response; do not substitute prices. |
Response503 · api_error | ConditionExact supplier route unavailable. | Next stepPreserve response for route repair. |
Response503 · result_pending_check | ConditionOriginal result requires checking; X-Should-Retry:false. | Next stepKeep original operation_id/id and read its status; never repeat original POST. |
Response404 · not_found | ConditionBatch status operation absent or owned by another identity. | Next stepUse exact original id and same user/key/group. |
Ailoot Ultra anthropic/claude-haiku-5.5:batch
Use the method paths shown above.