Fifty-four models. One key.

xAI Grok 4.7 — reasoning chat, 128K context. Provisional pricing based on Azure Grok 4.6: $2.50/MTok input, $7.50/MTok output, $0.625/MTok cached input. Reasoning tokens are billed as output.

DeepSeek V4 Pro — flagship of the V4 generation, frontier open-source reasoning, 1M context. Cached input $0.145/MTok — one twelfth of fresh.

DeepSeek V4 Flash — the cheap, fast V4 variant for high-throughput work, 1M context. Cached input $0.028/MTok — one seventh of fresh.

DeepSeek V4.1 Flash — reasoning and tool use, with 128K context. $0.46875/MTok input, $1.875/MTok output, $0.010/MTok cached input. Preview; Azure inference retires 2026-12-15.

DeepSeek V3.2 Speciale — reasoning-tuned V3.2, stronger on maths and competition-style problems.

DeepSeek V3.2 — frontier open-source general-purpose, 164K context. Reasoning lives inside the span. Pick for single-shot tasks.

Meta Llama 4 Maverick — first Llama 4 generation, 1M context, 12 languages, vision+text.

Meta Llama 3.3 70B Instruct — strong general-purpose with tool use.

Mistral Medium 3.5 — European mid-tier: image and PDF input, 18 languages, strong instruction following.

Mistral Large 3 — flagship European model, strong instruction following.

Microsoft GPT-OSS 120B — open-weight reasoning model.

Cohere Command-A Plus — upgraded Command-A: multilingual, agentic RAG and tool-calling, 128K context, image input.

Cohere Command-A — multilingual (10 languages), strong RAG and tool-calling.

xAI Grok 4.3 — latest frontier Grok: reasoning, tool use, image input, 200K context. Cached input $0.20/MTok — one sixth of fresh.

xAI Grok 4.1 Fast — fast Grok variant with reasoning + tool use.

Microsoft Phi-4 Mini — small, cheap reasoning model with thinking traces.

Moonshot AI Kimi K2.7 Code — coding-specialised K2 with thinking traces, 262K context, image input. Cached input $0.19/MTok — one fifth of fresh.

Moonshot AI Kimi K2.6 — latest frontier reasoning with thinking traces.

Moonshot AI Kimi K2.5 — strong reasoning model with thinking traces.

OpenAI GPT-6 Sol — coding and agentic workhorse, 1.05M context, image input. Chat function tools use reasoning_effort=none; use /v1/responses for tools with reasoning. Cached input $0.20/MTok — one tenth of fresh.

OpenAI GPT-6 Luna — efficient model for focused, high-volume tasks, 1.05M context, image input. Chat function tools use reasoning_effort=none; use /v1/responses for tools with reasoning. Cached input $0.01/MTok — one tenth of fresh.

OpenAI GPT-6 Astra — frontier reasoning and agentic model, 1.05M context, image input. Function tools are available on /v1/responses only; /v1/chat/completions rejects them for this model. Cached input $1.00/MTok — one tenth of fresh.

OpenAI GPT-5.6 Sol — flagship coding + agentic workhorse, 1.05M context. Sol Ultra mode via reasoning_effort=high. Cached input $0.50/MTok — one tenth of fresh.

OpenAI GPT-5.6 Terra — balanced tier, 1.05M context, everyday coding + reasoning at half Sol's price. Cached input $0.25/MTok — one tenth of fresh.

OpenAI GPT-5.6 Luna — fastest and cheapest GPT-5.6 variant, 400K context, high-throughput chat. Cached input $0.10/MTok — one tenth of fresh.

OpenAI GPT-5.5 — flagship, 1.05M context, image+text reasoning. Cached input $0.50/MTok — one tenth of fresh.

OpenAI GPT-5.4 — production workhorse, 1.05M context, strong agentic + long-doc. Cached input $0.25/MTok — one tenth of fresh.

OpenAI GPT-5.4 Mini — drop-in mid-tier, 400K context, reasoning. Cached input $0.075/MTok — one tenth of fresh.

OpenAI GPT-5.2 — improved reasoning and speed. Cached input $0.25/MTok — one tenth of fresh.

OpenAI GPT-5 — flagship proprietary model (first GPT-5 release). Cached input $0.25/MTok — one tenth of fresh.

Qwen 3 8B — fast multilingual, self-hosted on our own GPU fleet. Retiring 2026-10-23 — use any chat model, e.g. gpt-5.4-mini or llama-3.3-70b. Details.

Qwen 2.5 Coder 7B — code-specialised self-hosted. Retiring 2026-10-23 — use any chat model, e.g. gpt-5.4-mini or deepseek-v4-flash. Details.

Mistral Nemo 12B — strong reasoning, self-hosted on our own GPU fleet. Retiring 2026-10-23 — use mistral-medium-3.5 or any chat model. Details.

Google Gemma 3 27B — strong vision + text, self-hosted on our own GPU fleet. Retiring 2026-10-23 — use any chat model, e.g. gpt-5.4-mini or llama-3.3-70b. Details.

Zhipu AI GLM-4 9B — bilingual EN/ZH chat, self-hosted. Retiring 2026-10-23 — use any chat model, e.g. gpt-5.4-mini or deepseek-v4-flash. Details.

GPT-Image-2.5 Flare — text-to-image with low, medium, high, extra-high and max quality. Billed by actual tokens: $6.25/MTok text input and $37.50/MTok image output; size and quality affect the total.

GPT-Image-2.5 Sunburst — text-to-image with low, medium, high, extra-high and max quality. Billed by actual tokens: $6.25/MTok text input and $37.50/MTok image output; size and quality affect the total.

OpenAI GPT-Image-2 — flagship image generation, photoreal + editorial. Strong typography rendering.

OpenAI GPT-Image-1 Mini — cheap image generation, good for thumbnails and batch.

Black Forest Labs FLUX.2 Pro — open frontier image gen, strong typography + composition.

OpenAI Sora 2 — text-to-video. Async: POST /v1/videos → poll → fetch MP4. Sizes 720×1280, 1280×720, 1024×1792, 1792×1024.

Full-duplex voice: listens and speaks continuously over /v1/live/sessions, with ten voices and optional function-tool delegation. Uses the GPT-Live protocol. $0.09375/min of session clock, billed per second; delegated text-model tokens are additional. This price uses a provisional Azure cost basis. GPT-Live guide.

OpenAI Realtime API-compatible speech-to-speech over WebSocket at /v1/realtime — ~250 ms to first audio. Pair it with any chat model via ?model=. Server-side VAD with barge-in, automatic language detection across ten languages, transcript events both directions. ≈$0.0133/min typical incl. chat tokens. Realtime docs. Retiring 2026-10-23 — use kataleptic-realtime-hd or gpt-realtime-2.1-mini. Details.

Premium voice tier: same /v1/realtime WebSocket, served by Azure Voice Live — 600+ studio-grade HD neural voices, deep noise suppression, echo cancellation, semantic turn detection. Accepts G.711 (μ-law/A-law) for telephony. Exact transcripts, ~1.2 s to first audio. ≈$0.03/min typical. Realtime docs.

Latest native speech-to-speech tier: OpenAI gpt-realtime-2.1 behind the same WebSocket — successor to gpt-realtime-2 with better instruction following and turn-taking at the same token price, plus image input. Accepts G.711 for telephony. Transcripts are model approximations. ≈$0.07/min typical for voice; image input bills separately at $6.25/MTok. Realtime docs.

Volume tier for native speech-to-speech: roughly a third the audio-token cost of gpt-realtime-2.1, same protocol, voices and telephony codecs — sized for call handling. ≈$0.02/min typical for voice; image input bills separately at $1.00/MTok. Realtime docs.

Native speech-to-speech tier: OpenAI gpt-realtime-2 behind the same WebSocket — best-in-class prosody, hears tone rather than just words. ~1.0 s to first audio; accepts G.711 for telephony. Transcripts are model approximations. ≈$0.07/min typical for voice; image input bills separately at $6.25/MTok. Realtime docs.

Piper neural text-to-speech in ten languages (EN, DE, FR, ES, NL, SV, DA, IT, FI, RU). Streams PCM16; the voice behind /v1/realtime. Self-hosted. Retiring 2026-10-23 — use the HD or native realtime voices. Details.

OpenAI Whisper Large V3 Turbo — V3 distilled to ~6× faster decoding with near-identical WER. Self-hosted on our own GPU fleet. Retiring 2026-10-23 — use gpt-4o-mini-transcribe, or whisper-large-v2 for timestamps. Details.

Real-time streaming ASR over WebSocket. Deepgram-shape protocol at /v1/listen. Same Turbo model, VAD-segmented, partial + final transcripts. Retiring 2026-10-23 — use gpt-live-transcribe. Details.

NVIDIA NeMo Parakeet-TDT-0.6B v2 — English-only, SOTA Open-ASR-Leaderboard WER, VAD-segmented streaming. Self-hosted. Retiring 2026-10-23 — use gpt-live-transcribe. Details.

OpenAI GPT-4o Mini Transcribe — the cheapest file transcription at /v1/audio/transcriptions, for bulk and background audio. Returns json or text. $0.0045/min.

OpenAI GPT Transcribe — current-generation file transcription at /v1/audio/transcriptions, higher accuracy than GPT-4o Transcribe at the same per-minute rate. $0.006/min.

OpenAI Whisper large-v2 — file transcription with timestamps: verbose_json with word and segment timings, srt and vtt, for subtitles. $0.0097/min. Rate-limited to about 3 requests per minute; use gpt-4o-mini-transcribe for volume. Retires 2026-12-15; after that date no model on Kataleptic returns timestamps or subtitles.

OpenAI GPT Live Transcribe — streaming speech-to-text over WebSocket at /v1/realtime?intent=transcription: stream audio in, get partial and final transcripts back. The default transcription-session model. $0.02125/min. Streaming transcription docs.

OpenAI GPT Realtime Whisper — the second streaming speech-to-text model on the same transcription session at /v1/realtime?intent=transcription. $0.02125/min. Streaming transcription docs.

OpenAI GPT-4o Transcribe — higher-accuracy ASR with semantic context.

OpenAI GPT-4o Transcribe with speaker diarization — multi-speaker meetings.

High-quality multilingual text embeddings, self-hosted on our own GPU fleet. Retiring 2026-10-23 — use no replacement; no embedding model is offered after that date. Details.