# IronStratum

> Docs for the IronStratum inference API. Start at the quickstart; the pricing page is generated from the live catalog.

- / — OpenAI-compatible inference API. Same response shape on every request, a prepaid wallet with hard limits, and an API that only ever grows.
- /models — Every model on one metered API: chat, speech, transcription, embeddings, rerank, parsing, each row's state, one key, one wallet.
- /chat — One OpenAI-compatible LLM API for ten open-weight chat models from Qwen, DeepSeek, GLM, Gemma, MiniMax, Muse and Ornith. Metered per token.
- /tts — Four open-weight TTS models, Kokoro to Audio8, on one OpenAI-compatible speech endpoint, billed by the character from a prepaid wallet.
- /stt — Two open-weight speech to text API routes, batch and streaming, one meter: audio-time billing, the diarization question, how to choose.
- /embeddings — What an embeddings API does, the jobs it serves from semantic search to RAG, and how its route meters tokens against a prepaid balance.
- /rerank — What a rerank API does to a retrieved shortlist, when cross-encoders beat vector similarity, and how the route bills one search per call.
- /parsing — What a document parsing API does, the text-out, structured-parse, and archival-fidelity routes, and where each route stands.
- /pricing — How API pricing works here: five metering units, a cost report on every response, and a prepaid wallet that halts spend at zero.
- /models/qwen3.8-27b — Qwen 3.8 27B hosted: the full 262,144-token window, a per-request thinking switch, metered per token, from $0.35/1M input tokens.
- /models/qwen3.6-35b — Call Qwen3.6-35B-A3B on an OpenAI-compatible chat API: thinking control, 131,072-token served window, from $0.11/1M input tokens.
- /models/minimax-m2.7 — MiniMax M2.7 as a metered API: 196,608-token served window, tool calling, always-on thinking, OpenAI compatible. from $0.24/1M input tokens.
- /models/muse-glimmer-30b — Muse Glimmer 30B from Meta: agentic chat API with tool calling, image input, and a 131,072-token window. from $0.28/1M input tokens.
- /models/glm-5.3-flash — GLM 5.3 Flash from Z.ai: 320B/18B multimodal chat with a 1M window on IronStratum. from $0.11/1M input tokens.
- /models/deepseek-v4-flash-0731 — DeepSeek V4 Flash 0731 hosted: the July 31 GA snapshot, 1M context, thinking control, metered per token, from $0.15/1M input tokens.
- /models/gemma-4-31b-it — Gemma 4 31B IT hosted: the instruct variant of Google's open 31B, image input, thinking control, from $0.22/1M input tokens.
- /models/deepseek-v4-pro — DeepSeek V4 Pro hosted: the 0813 GA open weights, 1M context, three thinking efforts, metered per token, from $1.13/1M input tokens.
- /docs — Guides and API reference for the IronStratum inference API.
- /docs/quickstart — From zero to your first generated token in five steps: create a key, export two environment variables, and call the API with curl.
- /docs/pricing — Prices are generated from the live catalog: this table is what the platform charges, to the token, with nothing else on the bill.
- /docs/providers — IronStratum is an inference platform: we run and relay AI models behind one OpenAI-compatible API at `api.ironstratum.com`. Point any OpenAI SDK at it, create a key in the console, and call: no adapter layer, no dialect to learn.
- /docs/sessions — Server-side conversation continuity: send only the new turn, we assemble the context. Opt-in per request with one field, `session_id`, and every session-mode tool (list, inspect, cost, delete) lives under `/v1/conversations`.
- /docs/idempotency — Retries are safe here: send an `Idempotency-Key` header, and the same logical request returns the same logical response: no duplicate generation, no double billing. We shipped this for inference because nobody else has it, including OpenAI.
- /docs/streaming — Set `"stream": true` and the response is a server-sent-events (SSE) stream of chat completion chunks: same envelope, token by token.
- /docs/errors — Every error is one envelope: parse it once, handle it everywhere:
- /docs/rate-posture — Honest current state, no marketing fog: what limits exist today, what is reserved for later, and where the numbers live.
- /docs/beta-commitments — Five commitments, in the words we shipped them. Each maps to a community pain we designed against; this page is the contract you hold us to.
- /docs/real-time-speaker-diarization — Real-time speaker diarization answers one question while the audio is still flowing: who spoke when. This guide walks the four-step pipeline and shows why clustering is the hard part, explains why offline diarization falls apart on live crosstalk, teaches you to read the DER and cpWER numbers vendors quote, states the session-length limits almost nobody publishes, and works the build-or-buy math, using measured streaming latency from this platform's transcription lane and an honest account of where speaker labels stand today.
- /docs/merged-cell-pdf-table-extraction — Merged cells are the part of PDF table extraction that open-source tools have declined to solve: pdfplumber returns None cells, PyMuPDF duplicates values and invents column names, Camelot recovers some merged headers and produces phantom columns elsewhere, and the feature requests have sat open for years. This guide shows each failure on a test table you can rebuild with one script, using outputs captured from the real tools, then walks the routes that work: the text-versus-scanned fork, the repair step and the values it silently fabricates, the Excel conversion paths, and the per-page economics of repairing it yourself versus routing documents to a parsing service.
- /docs/avoid-api-bill-shock — API bill shock is what happens when usage-based charges run ahead of you: a leaked key, a retry loop, or a swarm of agent calls spends money faster than you notice, and the invoice arrives before the warning does. This guide breaks down the five ways bills blow up, compares the spend controls the major API providers actually ship, and shows how a prepaid wallet turns the worst case into a top-up instead of a debt.
- /docs/tts-local-vs-api-break-even — Local TTS versus a hosted API is a volume question, and volume questions have answers in arithmetic: this guide gives the crossover formula in characters per day, works it across two real GPU tiers at September 2026 street prices and the official United States electricity price, and shows the latency and long-text physics that stay identical on both sides of the choice, so the decision lands on your numbers rather than on someone else's rig.
- /docs/pdf-to-markdown-for-rag — The reliable way to turn a PDF corpus into RAG input is to convert to markdown first, then chunk the markdown by its headings, and this guide shows that path with the artifacts intact: a three-page test PDF run through three local parsers with every output quoted verbatim, a comparison table with dated cost per page from official rate cards, a runnable retrieval check that scores each parsing choice against a fixed question set, and a clear line for when running your own parsing stack stops paying.
- /docs/openai-compatible-api-guide — An OpenAI-compatible API is a wire format agreement, not a relationship with OpenAI: the same chat completions request and response shapes, the same bearer-key authentication, and the same streaming conventions, served by somebody else. This guide defines the phrase precisely, shows the two-line swap that moves any OpenAI SDK onto this platform, publishes the endpoint by feature compatibility matrix that no ranking page carries, lists the exact model strings to type, and gives you a five-minute test that proves what an endpoint really accepts, including the failure it returns when you send a parameter it does not know.
- /docs/long-audio-text-streaming-chunking — Long audio breaks naive transcription setups in three places: the file cap on a single request, the streaming session cap almost nobody publishes, and the client that tries to hold hours of audio in memory. This guide puts the duration and session limits of nine speech APIs and this platform in one dated table, answers the 30-minute speed question with a measured real-time factor instead of a human-typist ratio, gives you runnable chunking code for batch and streaming. The same spine then covers text output: model output caps, the continuation pattern, and the character caps on speech synthesis.
- /docs/ornith-local-vs-hosted — Running an open model on your own GPU or renting it per token is a volume question, and Ornith 1.5 is the worked example this guide uses to answer it: the family map across the 9B dense and 35B-A3B mixture-of-experts members, what each one needs in VRAM at each quantization, how fast it runs on named consumer cards including this platform's own dated measured baseline, which providers serve it hosted and at what published rates, and the crossover arithmetic that turns your electricity price and your street-price card into the daily token volume where local stops being cheaper.
- /docs/muse-connector-api-cost — Meta opened Muse connectors to developers on 2026-09-18: your service stops waiting to be installed and starts being reached by an agent when a person's task needs it. This guide lays out what Meta has actually published about the program, what an agent-facing API needs to have ready, where the metered model spend sits in a connector's backend, and how a prepaid wallet turns the worst backend bill into a top-up instead of a debt. It also separates the Muse agent platform from the Muse Glimmer open-weight model served on this platform, because the shared name is already tangling searches.
- /docs/muse-glimmer-local-vs-hosted — Muse Glimmer 30B is the open-weight agentic model Meta Superintelligence Labs distilled from Muse Spark, and because it is open weights you can run it on your own card or call it as a metered API. This guide gives the memory ladder from the 18 GB community-quant floor to the 60 GB full-precision build, the two local gotchas the listings do not advertise (a non-JSON native tool-call format, and reasoning controlled by a system-prompt phrase), the hosted rates on this platform and on the marketplace beside it, and the crossover arithmetic that tells you at what daily volume your own card wins.
- /docs/changelog — Additive changes to the public API, newest first. Per the versioning policy (`/v1` is additive-only, forever), nothing here ever removes or retypes a field: new fields, params, and endpoints appear; clients must tolerate additions. Dates are UTC.
- /docs/reference — Every API operation, machine-derived from openapi.yaml.
- /docs/reference/create-chat-completion — POST /v1/chat/completions
- /docs/reference/list-models — GET /v1/models
- /docs/reference/retrieve-model — GET /v1/models/{alias}
- /docs/reference/health — GET /v1/health
- /docs/reference/retrieve-credits — GET /v1/credits
- /docs/reference/query-hf-billing-batch — POST /v1/hf/billing
- /docs/reference/list-conversations — GET /v1/conversations
- /docs/reference/delete-all-conversations — DELETE /v1/conversations
- /docs/reference/retrieve-conversation — GET /v1/conversations/{id}
- /docs/reference/delete-conversation — DELETE /v1/conversations/{id}
- /docs/reference/create-embeddings — POST /v1/embeddings
- /docs/reference/create-speech — POST /v1/audio/speech
- /docs/reference/list-voices — GET /v1/audio/voices
- /docs/reference/create-transcription — POST /v1/audio/transcriptions
- /docs/reference/rerank-documents — POST /v1/rerank
- /docs/reference/ocr-document — POST /v1/ocr
- /docs/reference/ocr-job-create — POST /v1/ocr/jobs
- /docs/reference/ocr-job-status — GET /v1/ocr/jobs/{id}
- /docs/reference/stream-transcription — GET /v1/audio/transcriptions/stream
- /privacy — Last updated: [publication date]
- /terms — Last updated: [publication date]

## Models

- qwen3.8-27b — 262K context — $0.35/1M in · $2.55/1M out · $0.105/1M cached
- qwen3.6-35b — 131K context — $0.11/1M in · $0.8/1M out · $0.044/1M cached
- minimax-m2.7 — 197K context — $0.24/1M in · $0.95/1M out · $0.072/1M cached
- muse-glimmer-30b — 131K context — $0.28/1M in · $1.2/1M out · $0.084/1M cached
- ornith-1.5-35b — 100K context — $0.35/1M in · $2.55/1M out · $0.105/1M cached
- ornith-1.5-9b — 100K context — $0.1/1M in · $0.3/1M out · $0.03/1M cached
- glm-5.3-flash — 1049K context — $0.11/1M in · $0.35/1M out · $0.033/1M cached
- deepseek-v4-flash-0731 — 1311K context — $0.15/1M in · $0.42/1M out · $0.045/1M cached
- gemma-4-31b-it — 262K context — $0.22/1M in · $0.49/1M out · $0.066/1M cached
- deepseek-v4-pro — 1000K context — $1.13/1M in · $2.21/1M out · $0.339/1M cached
- chatterbox-tts — $25/1M chars
- kokoro-tts — $15/1M chars
- pocket-tts — $16/1M chars
- audio8-tts — $8/1M chars
- hayamimi-stt — $0.6/audio-hr
- bge-m3 — $0.05/1M tokens
- bge-reranker-v2-m3 — $1.5/1k searches
- whisper — $0.25/audio-hr