{ "schema_version": "eagleai/agents.json/2026-08-20", "name": "EagleAI Model API", "description": "Two frontier language models sold two ways from one domain. Ten flat-price endpoints where the price is per request and does not vary with response length, differing only in how large a request each accepts; and one uncapped per-token surface where the same two deployments are billed by the token over OpenAI Chat Completions. The fields at this level describe the flat-price endpoints; per_token_surface describes the other one.", "documentation_url": "https://eagleai.online/llms.txt", "privacy_policy_url": "https://eagleai.online/privacy.html", "terms_of_service_url": "https://eagleai.online/terms.html", "upstream_disclosure": "EagleAI resells access to these models. We do not build, own or train them, and the platform hosting our deployments is not named in this document - that is a commercial arrangement, not a fact about your data. We will name it in writing to any customer who asks, and to anyone who needs it for a data-protection assessment. What the arrangement means for request content is in data_handling and in the privacy notice, in full, and none of it changes with the name.", "provider": { "name": "EagleAI", "url": "https://eagleai.online", "contact": "mailto:sales@eagleai.online" }, "base_url": "https://eagleai.online", "authentication": { "type": "api_key", "in": "header", "header": "X-API-Key", "status_when_header_missing": 401, "status_when_key_invalid": 403 }, "pricing": { "unit": "request", "currency": "USD", "model": "flat_per_request", "notes": "One price per endpoint, charged per accepted request regardless of how many tokens the answer uses. A request refused for exceeding the endpoint's limits (400 or 413) is not billed." }, "purchase_channels": { "apihub": { "type": "marketplace", "settlement": "prepaid USDC on Base", "account_required_with_provider": false, "services": { "gpt-5.6-sol": "https://apihub.io/marketplace/eagleai-gpt56-sol", "claude-opus-5": "https://apihub.io/marketplace/eagleai-opus5" } }, "direct": { "type": "contact", "contact": "mailto:sales@eagleai.online", "notes": "Direct keys cover every endpoint, including the eight sizes not yet listed on the marketplace." } }, "error_codes": { "400": "unknown model name, or malformed body", "401": "no X-API-Key header", "403": "X-API-Key not recognised", "413": "request body or prompt above this endpoint's limit", "429": "over the per-minute request rate or token budget; Retry-After is sent in seconds" }, "models": [ { "id": "claude-opus-5", "vendor": "EagleAI", "wire_format": "messages", "openapi": "https://eagleai.online/api/v1/opus5/openapi.json", "request_body": { "required": [ "model", "messages", "max_tokens" ], "optional": [ "stream" ], "model_enum": [ "claude-opus-5" ] }, "endpoints": [ { "tier": "micro", "method": "POST", "path": "/api/v1/opus5/micro/messages", "url": "https://eagleai.online/api/v1/opus5/micro/messages", "price_usd": 0.0001, "max_input_tokens": 1000, "max_output_tokens": 1400, "reasoning_effort": "low", "requests_per_minute": 5, "purchase": "apihub", "below_cost": true, "min_output_tokens": 1200 }, { "tier": "nano", "method": "POST", "path": "/api/v1/opus5/nano/messages", "url": "https://eagleai.online/api/v1/opus5/nano/messages", "price_usd": 0.0009, "max_input_tokens": 2000, "max_output_tokens": 2000, "reasoning_effort": "low", "requests_per_minute": 3, "purchase": "apihub", "below_cost": true, "min_output_tokens": 1200 }, { "tier": "small", "method": "POST", "path": "/api/v1/opus5/small/messages", "url": "https://eagleai.online/api/v1/opus5/small/messages", "price_usd": 0.0849, "max_input_tokens": 3000, "max_output_tokens": 2400, "reasoning_effort": "low", "requests_per_minute": 2, "purchase": "direct", "below_cost": false, "min_output_tokens": 1200 }, { "tier": "medium", "method": "POST", "path": "/api/v1/opus5/medium/messages", "url": "https://eagleai.online/api/v1/opus5/medium/messages", "price_usd": 0.1009, "max_input_tokens": 4000, "max_output_tokens": 2800, "reasoning_effort": "low", "requests_per_minute": 2, "purchase": "direct", "below_cost": false, "min_output_tokens": 1200 }, { "tier": "large", "method": "POST", "path": "/api/v1/opus5/messages", "url": "https://eagleai.online/api/v1/opus5/messages", "price_usd": 0.111, "max_input_tokens": 4000, "max_output_tokens": 3000, "reasoning_effort": "medium", "requests_per_minute": 5, "purchase": "apihub", "below_cost": false, "min_output_tokens": 1200 } ] }, { "id": "gpt-5.6-sol", "vendor": "EagleAI", "wire_format": "openai.responses", "openapi": "https://eagleai.online/api/v1/openapi.json", "request_body": { "required": [ "model", "input" ], "optional": [ "stream", "max_output_tokens" ], "model_enum": [ "gpt-5.6-sol" ] }, "endpoints": [ { "tier": "micro", "method": "POST", "path": "/api/v1/micro/responses", "url": "https://eagleai.online/api/v1/micro/responses", "price_usd": 0.0001, "max_input_tokens": 150, "max_output_tokens": 80, "reasoning_effort": "none", "requests_per_minute": 60, "purchase": "apihub", "below_cost": true }, { "tier": "nano", "method": "POST", "path": "/api/v1/nano/responses", "url": "https://eagleai.online/api/v1/nano/responses", "price_usd": 0.0009, "max_input_tokens": 500, "max_output_tokens": 250, "reasoning_effort": "none", "requests_per_minute": 20, "purchase": "apihub", "below_cost": true }, { "tier": "small", "method": "POST", "path": "/api/v1/small/responses", "url": "https://eagleai.online/api/v1/small/responses", "price_usd": 0.0229, "max_input_tokens": 1000, "max_output_tokens": 500, "reasoning_effort": "none", "requests_per_minute": 10, "purchase": "apihub", "below_cost": false }, { "tier": "medium", "method": "POST", "path": "/api/v1/medium/responses", "url": "https://eagleai.online/api/v1/medium/responses", "price_usd": 0.0629, "max_input_tokens": 2100, "max_output_tokens": 1500, "reasoning_effort": "none", "requests_per_minute": 5, "purchase": "apihub", "below_cost": false }, { "tier": "large", "method": "POST", "path": "/api/v1/responses", "url": "https://eagleai.online/api/v1/responses", "price_usd": 0.1469, "max_input_tokens": 2100, "max_output_tokens": 4000, "reasoning_effort": "high", "requests_per_minute": 163, "purchase": "apihub", "below_cost": false } ] } ], "per_token_surface": { "name": "EagleAI Model API — per-token", "description": "The same two deployments, billed by the token rather than by the request, on a second base URL. None of the flat-price size caps apply: a caller may fill the whole context window and ask for the whole output ceiling, because on a per-token meter a large request is revenue rather than a fixed-price loss. Built to be resold by a router that invoices per token after the fact.", "documentation_url": "https://eagleai.online/llms.txt", "privacy_policy_url": "https://eagleai.online/privacy.html", "terms_of_service_url": "https://eagleai.online/terms.html", "base_url": "https://eagleai.online/api/openrouter/v1", "catalog_url": "https://eagleai.online/api/openrouter/v1/models", "wire_format": "openai.chat_completions", "wire_format_notes": "OpenAI Chat Completions in and out on both models. claude-opus-5 is translated to a messages-style API in front of the model and back again, so a client only ever speaks the one protocol.", "authentication": { "type": "api_key", "in": "header", "header": "Authorization", "scheme": "Bearer", "alternate_header": "X-API-Key", "status_when_header_missing": 401, "status_when_key_invalid": 401, "status_when_key_expired": 401 }, "pricing": { "unit": "token", "currency": "USD", "model": "per_token", "rates_url": "https://eagleai.online/api/openrouter/v1/models", "notes": "Billed per input token and per output token. The rates are deliberately not restated here: they are rendered from the same table the invoice is computed from and published, without a key, at rates_url. A discounted cached-input rate appears there for eagleai/gpt-5.6-sol and only for it, because that upstream caches prompts on its route automatically while nothing on the Opus route ever sends a cache flag - a cached rate there could never be earned." }, "purchase_channels": { "openrouter": { "type": "marketplace", "settlement": "per token, invoiced after the fact", "account_required_with_provider": false, "notes": "Listed as a provider on OpenRouter. A key carrying a prepaid credit allowance is refused on this surface rather than metered: with no per-request size cap there is no worst case to reserve against, so the allowance could never be spent down." }, "direct": { "type": "contact", "contact": "mailto:sales@eagleai.online", "notes": "A direct key authenticates here as well, with the same two headers." } }, "error_codes": { "400": "unknown model name, invalid JSON, or a parameter this surface will not silently drop", "401": "missing, unrecognised or expired key", "403": "key not authorised for that model, or carrying a prepaid credit allowance", "413": "request body too large", "422": "body failed validation; the offending field is in error.param", "429": "over the per-minute request, token or concurrency limit; Retry-After is sent in seconds", "502": "upstream connection error or malformed upstream reply", "503": "the surface is switched off or not configured on this server", "504": "the upstream request timed out" }, "error_body_shape": "openai.error_object", "models": [ { "id": "eagleai/claude-opus-5", "vendor": "EagleAI", "wire_format": "openai.chat_completions", "upstream_wire_format": "messages", "request_body": { "required": [ "model", "messages" ], "optional": [ "stream", "max_tokens", "max_completion_tokens", "temperature", "stop", "tools", "tool_choice", "parallel_tool_calls" ], "model_enum": [ "eagleai/claude-opus-5" ] }, "input_modalities": [ "text" ], "supports_streaming": true, "supports_tools": true, "supports_reasoning": true, "reasoning_control": "off_switch_only", "reasoning_control_notes": "The model reasons by default and those tokens are billed as output tokens, so supports_reasoning is true. Depth cannot be raised: reasoning_effort and an explicit thinking budget are both refused upstream. It can be turned off, with reasoning false or {enabled: false} or {exclude: true}. Neither name is declared in supported_parameters, because the one shape that works is the one nobody routes traffic for, and declaring it would attract the shapes that 400. Read supports_reasoning as 'it reasons, and you may stop it', not as 'you can ask it to reason harder'.", "supports_structured_outputs": false, "unsupported_parameters_are_refused": true, "notes": "Structured outputs are not offered: the upstream refuses response_format on its messages route, so asking for one is a 400 from here naming the parameter rather than a 400 from an upstream the caller cannot see. Anything else this route cannot honour is refused the same way rather than dropped quietly - n other than 1, logprobs, top_logprobs, a non-empty logit_bias, a non-zero frequency_penalty or presence_penalty, seed, and unrecognised message content parts - with the offending field named in error.param. Sending any of them as null is accepted, since that is how an OpenAI client spells not set, as are frequency_penalty 0 and presence_penalty 0, which ask for what the upstream already does. parallel_tool_calls false is genuinely honoured, and user is accepted and forwarded as the upstream's metadata.user_id. This is a reasoning model and its sampling knobs are fixed: temperature is accepted only at its default of 1 and any other value is refused upstream, while top_p and top_k are refused at every value. stop is honoured, but a stop sequence consisting only of whitespace is refused. It reasons by default and those tokens are billed as output tokens. That can be switched off - reasoning false, reasoning {enabled: false} and reasoning {exclude: true} all reach the upstream as a disabled thinking block and all return 200 - but it cannot be turned up: reasoning_effort, reasoning {effort} and reasoning {max_tokens} are each refused upstream, so how hard it thinks when it does think is the deployment's own setting. Measured 2026-08-20, one prompt, three runs each: 197/182/184 output tokens with thinking left on against 137/141/141 with it off, so the switch is real and it moves the bill.", "endpoints": [ { "method": "POST", "path": "/api/openrouter/v1/chat/completions", "url": "https://eagleai.online/api/openrouter/v1/chat/completions", "max_input_tokens": 200000, "max_output_tokens": 64000, "requests_per_minute": 40, "concurrent_requests": 4, "tokens_per_minute": 40000, "purchase": "openrouter", "notes": "The two token figures are the model's own context window and output ceiling, not a cap this surface imposes - nothing here refuses a request for its token count, and a max_tokens above the ceiling is clamped to it rather than rejected. The three limits after them are the deployment's own provisioned quota rather than a number we chose, and all three are enforced before the upstream is contacted. The token budget is a ceiling on the deployment rather than an allowance handed to each key, so every caller of this model draws on the same minute: a request is refused once that minute is already spent, and never for being large. What a request spends against the minute is its measured prompt at admission, corrected to the real prompt-plus-completion total as soon as the upstream reports one, so the window trails only by what is in flight. Where the deployment declares a separate output allowance, completions are metered against that as well. Over any of the three is an immediate 429 with Retry-After. Nothing queues." } ] }, { "id": "eagleai/gpt-5.6-sol", "vendor": "EagleAI", "wire_format": "openai.chat_completions", "upstream_wire_format": "openai.chat_completions", "request_body": { "required": [ "model", "messages" ], "optional": [ "stream", "max_tokens", "max_completion_tokens", "temperature", "top_p", "seed", "tools", "tool_choice", "response_format" ], "model_enum": [ "eagleai/gpt-5.6-sol" ] }, "input_modalities": [ "text" ], "supports_streaming": true, "supports_tools": true, "supports_reasoning": true, "reasoning_control": "reasoning_effort", "reasoning_control_notes": "Depth is selectable per request with reasoning_effort, which this wire takes and forwards at none, low, medium, high or xhigh; minimal is refused by this deployment. The reasoning parameter is refused upstream as an unknown field, so it is not declared in supported_parameters; send reasoning_effort rather than reasoning.", "supports_structured_outputs": true, "unsupported_parameters_are_refused": false, "notes": "Forwarded to the upstream deployment as sent, so a field added to the Chat Completions API after this document was written works without a change here. max_tokens is normalised to max_completion_tokens before it goes upstream, because the deployment is a reasoning model and its v1 chat/completions surface takes the latter; either name may be sent. The model field in every reply, streamed or not, is the published id above rather than the internal deployment name. Being a reasoning model, its sampling knobs are fixed: temperature and top_p are accepted only at their default of 1 and any other value is refused upstream, and stop is refused at every value. Reasoning depth is set with reasoning_effort, which this wire takes and forwards unaltered at none, low, medium, high or xhigh; minimal is the one value this deployment refuses, with a 400 naming it. The reasoning parameter is refused upstream as an unknown field, so send reasoning_effort rather than reasoning.", "endpoints": [ { "method": "POST", "path": "/api/openrouter/v1/chat/completions", "url": "https://eagleai.online/api/openrouter/v1/chat/completions", "max_input_tokens": 272000, "max_output_tokens": 128000, "requests_per_minute": 1000, "concurrent_requests": 203, "tokens_per_minute": 1000000, "purchase": "openrouter", "notes": "The two token figures are the model's own context window and output ceiling, not a cap this surface imposes - nothing here refuses a request for its token count, and a max_tokens above the ceiling is clamped to it rather than rejected. The three limits after them are the deployment's own provisioned quota rather than a number we chose, and all three are enforced before the upstream is contacted. The token budget is a ceiling on the deployment rather than an allowance handed to each key, so every caller of this model draws on the same minute: a request is refused once that minute is already spent, and never for being large. What a request spends against the minute is its measured prompt at admission, corrected to the real prompt-plus-completion total as soon as the upstream reports one, so the window trails only by what is in flight. Where the deployment declares a separate output allowance, completions are metered against that as well. Over any of the three is an immediate 429 with Retry-After. Nothing queues." } ] } ] }, "data_handling": { "prompts_stored": false, "completions_stored": false, "prompts_logged": false, "used_for_training": false, "metering_record_retention_days": 365, "metering_record_max_fields": 33, "metering_record_contains_content": false, "metering_record_caller_identifier": "the first sixteen hex characters of the SHA-256 of the API key presented - stable and one-way, so pseudonymous rather than anonymous", "prepaid_credit_ledger": { "applies_to": "keys issued with a prepaid credit allowance, on the flat-price endpoints only", "purpose": "so that a fixed grant cannot be overspent by requests still in flight", "fields": [ "request_id", "customer", "reserved", "settled", "state", "opened_at", "closed_at" ], "contains_content": false, "retention_days": null, "notes": "Amounts and states only - no tokens, no model, no prompt, no completion. There is no retention window and we do not claim one: the only deletion in the code is an operator clearing the table by hand. A key with no allowance never opens this ledger, and the per-token surface refuses allowance-carrying keys outright rather than metering them." }, "upstream_processor": "A third-party cloud AI platform, not named here. See upstream_disclosure: we name it in writing to any customer who asks. It is a processor for the inference itself and nothing else - it is not given our metering records, our billing data, or the list of who holds a key.", "upstream_abuse_monitoring_retention_days": 30, "zero_data_retention": false, "notes": "Prompt and completion content is held in memory only for the duration of the request, on both surfaces. One metering record per request - identifiers, token counts, timings and money, 33 fields, no content - is kept for 365 days and then deleted by the server. The sweep runs on write and again whenever the per-token model catalogue is polled, so retention does not depend on new inference arriving to trigger it. Modified abuse monitoring (zero data retention) is not enabled on these deployments; see the privacy policy." } }