Kataleptic
Vol. 01 № 2026 · 04
Status — Operating

Inference, served clean. Frontier models, answered with conviction.

Kataleptic — from the Stoic kataleptikē phantasia, the impression that grasps the world with certainty — is an OpenAI-compatible inference endpoint serving fifty-four frontier open-source and proprietary models through one billing surface.

§ 01

One endpoint.

Fifty-four models, three backends, one OpenAI-compatible surface. No proxy hops, no SDK rewrites, no surprise rate limits.

  1. Catalogue Live

    Fifty-four curated models

    GPT-6 Astra/Sol/Luna, GPT-5.6 Sol/Terra/Luna, DeepSeek V4 Pro/Flash and V3.2, Llama 3.3/4-Maverick, Mistral Large 3 / Medium 3.5 / Nemo, GPT-5/5.2/5.4/5.4-mini/5.5, GPT-OSS 120B, Cohere Command-A(+), Phi-4 Mini Reasoning, Kimi K2.5/K2.6/K2.7-Code, Grok 4.3 and 4.1 Fast, Gemma 3 27B, GLM-4 9B, Qwen 3 8B, Qwen 2.5 Coder, plus realtime voice, transcription, image, video and embeddings.

  2. Sovereign Live

    Sovereign fleet on dedicated metal

    For models we self-host, inference runs on our own fleet of GPU servers in our own colocation, orchestrated by xinity-ai. Open-weight only, transparent stack, no upstream provider in the loop.

§ 02

A line of code,
then you're inferring.

~/kataleptic · cURL
$ curl https://api.kataleptic.com/v1/chat/completions \
  -H "Authorization: Bearer dg_…" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-3.3-70b",
    "messages": [
      { "role": "user",
        "content": "Compose a haiku about inference." }
    ],
    "stream": true
  }'

> data: {"id":"chatcmpl-…","provider":"duguet-ai",
> data:   "choices":[{"delta":{"content":"Quiet servers hum,"}}]}
> data: {"choices":[{"delta":{"content":" one token at a time—"}}]}
> data: {"choices":[{"delta":{"content":" the answer arrives."}}]}
> data: [DONE]
§ 03

The price list

Transparent per-million-token pricing. No top-up fee, no hidden request surcharge. Billed by the token, paid from prepaid credit. First $2 on us.

Where a model supports prompt caching, re-sent context bills at the cached rate shown beneath the input price — one tenth of fresh input, and we charge nothing extra to write the cache. Agents that resend a conversation every turn pay a fraction of the headline number.

Model Context Input / MTok Output / MTok Backend
— Open-source chat —
GLM-4 9B retiring 2026-10-23 glm4-9b128k$0.05$0.08Sovereign fleet · xinity
Qwen 3 8B retiring 2026-10-23 qwen3-8b32k$0.05$0.08Sovereign fleet · xinity
Qwen 2.5 Coder 7B code retiring 2026-10-23 qwen2.5-coder-7b32k$0.05$0.08Sovereign fleet · xinity
Mistral Nemo 12B retiring 2026-10-23 mistral-nemo-12b128k$0.10$0.15Sovereign fleet · xinity
Gemma 3 27B retiring 2026-10-23 gemma3-27b128k$0.15$0.20Sovereign fleet · xinity
Llama 3.3 70B llama-3.3-70b131k$0.30$0.40Azure MaaS
DeepSeek V3.2 deepseek-v3-2131k$0.40$0.60Azure MaaS
Mistral Large 3 mistral-large-3262k$0.40$1.20Azure MaaS
Cohere Command-A new cohere-command-a288k$2.50$10.00Azure MaaS
DeepSeek V4 Pro tools new deepseek-v4-pro1M$1.74$3.48Azure MaaS
DeepSeek V4 Flash tools new deepseek-v4-flash1M$0.19$0.51Azure MaaS
DeepSeek V4.1 Flash preview · retires 2026-12-15 deepseek-v4.1-flash1M$0.46875$0.010 cached$1.875Azure MaaS
Mistral Medium 3.5 vision new mistral-medium-3.5262k$1.50$7.50Azure MaaS
Cohere Command-A Plus new cohere-command-a-plus131k$0.80$3.20Azure MaaS
Llama 4 Maverick 17B vision new llama-4-maverick128k$0.50$1.50Azure MaaS
— Reasoning —
GPT-OSS 120B reasoning gpt-oss-120b131k$0.15$0.60Azure MaaS
DeepSeek V3.2 Speciale reasoning new deepseek-v3-2-speciale131k$0.58$1.68Azure MaaS
Kimi K2.7 Code reasoning new kimi-k2.7-code262k$0.95$4.00Azure MaaS
xAI Grok 4.3 reasoning new grok-4.3200k$1.25$2.50Azure MaaS
Kimi K2.5 reasoning kimi-k2.5262k$0.60$2.40Azure MaaS
Kimi K2.6 reasoning kimi-k2.6262k$0.60$2.40Azure MaaS
xAI Grok 4.1 Fast reasoning new grok-4-1-fast128k$0.20$0.50Azure MaaS
xAI Grok 4.7 provisional 4.6 pricing grok-4.7500k$2.50$0.625 cached$7.50Azure MaaS
Phi-4 Mini Reasoning reasoning new phi-4-mini-reasoning4k$0.10$0.30Azure MaaS
— Proprietary —
GPT-5 gpt-5400k$2.50$0.25 cached$10.00Azure OpenAI
GPT-5.2 gpt-5.2400k$2.50$0.25 cached$14.00Azure OpenAI
GPT-5.4 Mini vision new gpt-5.4-mini400k$0.75$0.075 cached$4.50Azure OpenAI
GPT-5.4 vision new gpt-5.41M$2.50$0.25 cached$15.00Azure OpenAI
GPT-5.5 vision flagship new gpt-5.51M$5.00$0.50 cached$30.00Azure OpenAI
GPT-5.6 Luna vision new gpt-5.6-luna400k$1.00$0.10 cached$6.00Azure OpenAI
GPT-5.6 Terra vision new gpt-5.6-terra1M$2.50$0.25 cached$15.00Azure OpenAI
GPT-5.6 Sol vision flagship new gpt-5.6-sol1M$5.00$0.50 cached$30.00Azure OpenAI
GPT-6 Luna vision new gpt-6-luna1.05M$0.10$0.01 cached$0.50Azure OpenAI
GPT-6 Sol vision new gpt-6-sol1.05M$2.00$0.20 cached$10.00Azure OpenAI
GPT-6 Astra vision flagship new gpt-6-astra1M$10.00$1.00 cached$50.00Azure OpenAI
— Embeddings —
Nomic Embed retiring 2026-10-23 nomic-embed8k$0.02—Sovereign fleet · xinity
— Image generation · flat rates at 1024×1024; 2.5 variants per MTok —
GPT-Image-1 Mini new gpt-image-1-mini—$0.011 / imageAzure OpenAI
FLUX.2 Pro new flux-2-pro—$0.040 / imageAzure MaaS
GPT-Image-2 flagship new gpt-image-2—$0.080 / image · high ×2Azure OpenAI
GPT-Image-2.5 Flare new gpt-image-2.5-flare—$6.25 / MTok text$37.50 / MTok imageAzure OpenAI
GPT-Image-2.5 Sunburst new gpt-image-2.5-sunburst—$6.25 / MTok text$37.50 / MTok imageAzure OpenAI
— Full-duplex voice · per minute of session clock —
GPT-Live 1 new gpt-live-1—$0.09375 / min + delegate tokensAzure OpenAI · provisional pricing
— Speech recognition · per minute of audio —
Whisper Large V3 Turbo new fastest retiring 2026-10-23 whisper-large-v3-turbo—$0.003 / minSovereign fleet · xinity
GPT-4o Transcribe new gpt-4o-transcribe—$0.006 / minAzure OpenAI
GPT-4o Transcribe (diarized) new gpt-4o-transcribe-diarize—$0.024 / minAzure OpenAI
GPT-4o Mini Transcribe new cheapest gpt-4o-mini-transcribe—$0.0045 / minAzure OpenAI
GPT Transcribe new gpt-transcribe—$0.006 / minAzure OpenAI
Whisper large-v2 timestamps retires 2026-12-15 whisper-large-v2—$0.0097 / minAzure OpenAI
GPT Live Transcribe streaming new gpt-live-transcribe—$0.02125 / minAzure OpenAI
GPT Realtime Whisper streaming new gpt-realtime-whisper—$0.02125 / minAzure OpenAI
— Video generation · per second of output, async job —
Sora 2 new sora-2—$0.20 / secAzure OpenAI

A note on price: we aren't chasing the absolute floor. We undercut the major aggregators by 20–40 % on most SKUs and don't charge a top-up fee.

§ Volume discounts Auto-applied · rolling 30-day window · no contracts
  • $0–$500 / moPay-as-you-golisted price
  • $500+ / moScale5 % off
  • $2,000+ / moVolume10 % off
  • $10,000+ / moEnterprise15 % off · talk to us
§ 04

The workshop

i.

From first principles.

We build the smallest piece that solves the actual problem. Fewer abstractions, more measurements, and a clear paper trail from input to output.

ii.

Open where it matters.

We prefer open weights and open formats; we contribute the generalisable parts back. Proprietary work earns its keep on results, not on lock-in.

iii.

Production-shaped.

Every release ships with throughput numbers, latency p50/p99, failure modes, and a runbook. If it cannot be operated, it is not finished.

Claim your API key.

$2 in free trial credit for eligible new inboxes, no card. Drop-in replacement for OpenAI or OpenRouter clients — point base URL at https://api.kataleptic.com/v1 and go.

Already have an account but lost the key? Recover by email →

One inbox, read by a human · [email protected]