Research

Latest model families, open evaluations, and developer resources — the same structure you use in the nav: Latest models and For developers.

Latest models

Frontier checkpoints, documented limits

Each release ships with benchmark tables, known failure modes, and recommended deployment profiles. Icons mirror the Research menu for quick visual scanning.

Fast · cost-efficient

ideasbd Compact

A small dense model for high-volume chat, classification, and extraction. Optimized for sub-200ms p95 on single-GPU inference. Ideal when latency and unit economics matter more than maximum reasoning depth.

  • Context: 32k tokens · batch-friendly serving
  • Strong on structured JSON and tool calling
  • Published MMLU / MT-Bench deltas vs. prior Compact generation

Flagship reasoning

ideasbd Core

Our primary frontier model for agents, long analysis, and multi-step planning. Trained with improved data filtering and refusal behavior; ships with red-team summary and system-card template.

  • Context: 128k · native JSON & function schemas
  • Benchmarks: reasoning suites + coding leaderboards
  • Pair with Console for policy & eval gates

Speech · TTS

ideasbd Voice

Low-latency text-to-speech and streaming transcription with speaker diarization hooks. Designed for IVR, accessibility, and meeting capture when paired with Audio Transcription in the product suite.

  • SSML-friendly prosody controls (limited beta)
  • Word-level timestamps for alignment with UI
  • Voice clone workflows require explicit enterprise review

Lean · edge-ready

ideasbd Edge

Distilled weights for on-device and VPC-constrained environments. Quantization recipes (INT8 / INT4) validated for accuracy regression thresholds you can enforce in CI.

  • Targets CPU + small GPU footprints
  • ONNX and GGUF export paths from Model Forge
  • Ideal for air-gapped or high-latency edge sites

See all models

Historical checkpoints, deprecated versions, and experimental heads. Use this list when pinning API versions or reproducing paper results.

For developers

Docs, API, and cookbooks

Everything here is mirrored from the Research mega-menu. Replace placeholder links with your real docs host when you go live.

Documentation

Conceptual guides: authentication, workspaces, rate limits, streaming, and error codes. Start with quickstart, then routing and retrieval patterns.

Quick links

Authentication (API keys + workspace scopes) · Streaming SSE · Webhooks for job completion · Regional endpoints

API reference

OpenAPI 3.1 schema and per-route examples for chat, embeddings, fine-tuning jobs, and batch inference. Versioned by date so you can pin /v2026-04-06/ paths in production.

Coverage

Request/response models · idempotency keys · pagination cursors · deprecated field warnings in headers

Cookbooks

Copy-paste notebooks and snippets: RAG with citations, human-in-the-loop tool use, eval harnesses, and cost-aware routing between Compact and Core.

Popular recipes

Grounded support bot · Contract clause extractor · CI linter with ideasbd patch suggestions

Publications & lab notes

Long-form writeups and external venues. Subscribe via our team for new releases.

Collaborate with us