{ "schema_version": 1, "fixture": false, "started_at": "2026-11-02T16:38:54.876Z", "finished_at": "2026-11-02T16:39:02.544Z", "endpoint": "https://api.excessionlabs.ai/v1", "runtime": "v26.12.1", "client_platform": "darwin/arm64", "concurrency": 1, "order": "Alternate standard then abliterated, repeated samples times; no separate warmup; same prompt every request.", "samples_per_model": 2, "request_template": { "messages": [ { "role": "user", "content": "Explain HTTP caching to a developer in exactly five short numbered sentences. Mention Cache-Control, ETag, conditional requests, freshness, and private responses. Answer directly without a preamble." } ], "temperature": 0, "max_tokens": 160, "stream": true, "stream_options": { "include_usage": true }, "reasoning": { "enabled": false }, "thinking": { "type": "disabled" } }, "timing_definition": "Client wall time from fetch invocation. first_token_ms is first nonempty content OR reasoning delta; first_content_ms excludes reasoning. Includes network, queueing and generation. Not provider-only latency.", "cost_definition": "Estimate using returned usage and publicly advertised catalog rates captured before run; not an invoice or observed charge. Cache counted only where usage reports it. No plan bonus applied.", "limitations": [ "Tiny sequential sample, one client/network and one short prompt; not a capacity, quality, long-context, or availability benchmark.", "No regional or competing-provider comparison. No p95 or SLA claim. Caches and routing may vary.", "Maximum 160 completion tokens per request; finish_reason length means truncation." ], "catalog": { "object": "list", "data": [ { "id": "deepseek/deepseek-v4.1-flash-1017", "object": "model", "created": 1793318400, "owned_by": "excession", "context_length": 1048576, "max_output_tokens": 131072, "pricing": { "currency": "USD", "input_per_million_tokens": "0.28", "cached_input_per_million_tokens": "0.07", "output_per_million_tokens": "1.00" } }, { "id": "deepseek-v4.1-flash-abliterated", "object": "model", "created": 1793318400, "owned_by": "excession", "context_length": 1048576, "max_output_tokens": 131072, "pricing": { "currency": "USD", "input_per_million_tokens": "0.28", "cached_input_per_million_tokens": "0.07", "output_per_million_tokens": "1.00" } } ] }, "summary": [ { "model": "deepseek/deepseek-v4.1-flash-1017", "requests": 2, "successful_streams": 2, "errors": 0, "median_first_token_ms": 904.15, "median_first_content_ms": 904.15, "median_total_ms": 2878.215, "usage_reported_requests": 2, "estimated_cost_usd_at_catalog_rates": 0.00034184000000000005 }, { "model": "deepseek-v4.1-flash-abliterated", "requests": 2, "successful_streams": 2, "errors": 0, "median_first_token_ms": 290.385, "median_first_content_ms": 290.385, "median_total_ms": 908.8050000000001, "usage_reported_requests": 2, "estimated_cost_usd_at_catalog_rates": 0.00026984000000000003 } ], "requests": [ { "model": "deepseek/deepseek-v4.1-flash-1017", "sample": 1, "started_at": "2026-11-02T16:38:54.969Z", "http_status": 200, "headers_ms": 1035.04, "first_token_ms": 1035.28, "first_content_ms": 1035.28, "total_ms": 3137.23, "completed": true, "finish_reason": "stop", "usage": { "prompt_tokens": 39, "completion_tokens": 160, "total_tokens": 199, "cost": 0, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "output": "1. HTTP caching stores responses so clients and intermediaries can reuse them without contacting the origin server every time.\n2. The `Cache-Control` header defines directives like `max-age` and `no-store` that control how and how long a response may be cached.\n3. Freshness determines whether a cached response can be reused directly, based on its age versus the lifetime set by `max-age` or `Expires`.\n4. When a cached response is stale, the client can send a conditional request using `If-None-Match` with the stored `ETag`, and the server replies `304 Not Modified` if it still matches.\n5. Mark responses with `Cache-Control: private` when they contain user-specific data so only the browser caches them, not shared proxies.", "reasoning_characters": 0, "error": null, "estimated_cost_usd_at_catalog_rates": 0.00017092000000000002 }, { "model": "deepseek-v4.1-flash-abliterated", "sample": 1, "started_at": "2026-11-02T16:38:58.107Z", "http_status": 200, "headers_ms": 319.98, "first_token_ms": 320.1, "first_content_ms": 320.1, "total_ms": 977.88, "completed": true, "finish_reason": "stop", "usage": { "prompt_tokens": 39, "completion_tokens": 124, "total_tokens": 163, "prompt_tokens_details": { "cached_tokens": 0 }, "prompt_cache_hit_tokens": 0, "prompt_cache_miss_tokens": 39, "cost": 0 }, "output": "1. HTTP caching stores responses so clients and intermediaries can reuse them without contacting the origin server.\n2. Cache-Control directives like max-age and no-store define how long a response is fresh and whether it may be cached at all.\n3. An ETag is a validator the server assigns to a representation, letting clients detect changes via conditional requests such as If-None-Match.\n4. Conditional requests return 304 Not Modified when the cached copy is still valid, saving bandwidth while confirming freshness.\n5. Mark responses private when they contain user-specific data so only the end client's cache stores them, not shared proxies.", "reasoning_characters": 0, "error": null, "estimated_cost_usd_at_catalog_rates": 0.00013492000000000002 }, { "model": "deepseek/deepseek-v4.1-flash-1017", "sample": 2, "started_at": "2026-11-02T16:38:59.085Z", "http_status": 200, "headers_ms": 772.92, "first_token_ms": 773.02, "first_content_ms": 773.02, "total_ms": 2619.2, "completed": true, "finish_reason": "length", "usage": { "prompt_tokens": 39, "completion_tokens": 160, "total_tokens": 199, "cost": 0, "prompt_tokens_details": { "cached_tokens": 0, "cache_write_tokens": 0, "audio_tokens": 0, "video_tokens": 0 }, "completion_tokens_details": { "reasoning_tokens": 0, "image_tokens": 0, "audio_tokens": 0 } }, "output": "1. HTTP caching stores responses so clients and intermediaries can reuse them without contacting the origin server every time.\n2. The `Cache-Control` header defines caching rules like `max-age` and `no-store`, controlling how long and where a response may be cached.\n3. Freshness determines whether a cached response can be reused directly, based on its age versus the lifetime set by `max-age` or `Expires`.\n4. When a response is stale, the client sends a conditional request using `ETag` (via `If-None-Match`) or `Last-Modified`, and the server replies `304 Not Modified` if unchanged.\n5. Mark responses with `Cache-Control: private` when they contain user-specific data, so only the browser caches them and shared proxies", "reasoning_characters": 0, "error": null, "estimated_cost_usd_at_catalog_rates": 0.00017092000000000002 }, { "model": "deepseek-v4.1-flash-abliterated", "sample": 2, "started_at": "2026-11-02T16:39:01.704Z", "http_status": 200, "headers_ms": 260.57, "first_token_ms": 260.67, "first_content_ms": 260.67, "total_ms": 839.73, "completed": true, "finish_reason": "stop", "usage": { "prompt_tokens": 39, "completion_tokens": 124, "total_tokens": 163, "prompt_tokens_details": { "cached_tokens": 0 }, "prompt_cache_hit_tokens": 0, "prompt_cache_miss_tokens": 39, "cost": 0 }, "output": "1. HTTP caching stores responses so clients and intermediaries can reuse them without contacting the origin server.\n2. Cache-Control directives like max-age and no-store define how long a response is fresh and whether it may be cached at all.\n3. An ETag is a validator the server assigns to a representation, letting clients detect changes via conditional requests such as If-None-Match.\n4. Conditional requests return 304 Not Modified when the cached copy is still valid, saving bandwidth while confirming freshness.\n5. Mark responses private when they contain user-specific data so only the end client's cache stores them, not shared proxies.", "reasoning_characters": 0, "error": null, "estimated_cost_usd_at_catalog_rates": 0.00013492000000000002 } ] }