Model catalogue

Sports-2

infersia/sports-2

·Infersia

Experimental A/B of the Sports-1 grounding stack on a different base model.

Experimental sibling of Sports-1: Infersia's proprietary sports grounding stack — live betting lines, player props, scores, standings, results and roster tools, the settlement rulebook and the betslip reader — on a 27-billion-parameter dense vision model. Dense means all 27B parameters compute on every token (Sports-1's mixture-of-experts activates 18B of its 320B), which is what the higher output price pays for. Multimodal (reads betslip screenshots). Here to be compared, not to replace Sports-1. 1M-token context (long-context requests are served text-only).

Sports-1

infersia/sports-1

·Infersia

Infersia's proprietary sports grounding stack on a 320-billion-parameter sparse mixture-of-experts base — 18B parameters active per token, which is why big-model quality prices this low. Live sports data the gateway fetches itself, mid-response: betting lines — moneyline, spread and totals — and player props (anytime and first touchdown scorer, passing, rushing and receiving yards, points, rebounds, assists, home runs, strikeouts, goals and more) from named bookmakers for every major sport (NFL, NBA, MLB, NHL, college, UFC, tennis, golf, cricket, AFL, NRL and soccer leagues worldwide), roster lookups for the US leagues, plus deep soccer coverage: fixture odds, league tables, fixtures and results, squads and transfers. Send an ordinary chat completion and grounding is automatic — no tool loop to implement, no data-provider key to hold. Prices are quoted from a named bookmaker in the format its market actually uses, never an average and never converted. Bring your own tools and they take precedence. 1M-token context window.

DeepSeek V4.1 Flash

deepseek/deepseek-v4.1-flash

·DeepSeek

DeepSeek V4.1 Flash is a sparse mixture-of-experts model and the first built on DeepSeek's Causal Encoder-Decoder architecture, activating 8B parameters on input and 16B on output. It accepts text and images, holds a 1,048,576-token context, and supports tool calling, structured output and reasoning control. The asymmetric activation makes it unusually cheap on long inputs, which suits whole-repository code work, long document synthesis and agent runs that accumulate a large history.

DeepSeek V4 Flash

deepseek/deepseek-v4-flash-0731

·DeepSeek

DeepSeek V4 Flash is a 284B mixture-of-experts model activating roughly 13B parameters per token, with a 1M context window. It uses compressed latent attention, so a very long prompt costs far less memory than its length suggests. Built for work that needs a whole corpus in one pass — repository-wide code analysis, long document synthesis and multi-step reasoning over large inputs.

DeepSeek V4 Pro

deepseek/deepseek-v4-pro-0813

·DeepSeek

The GA release of DeepSeek V4 Pro: a large mixture-of-experts model with a one-million-token context window and strong tool-calling and agentic behaviour.

Inkling Small

thinkingmachines/inkling-small

·Thinking Machines Lab

Natively multimodal 276B mixture-of-experts model from Thinking Machines Lab, activating roughly 12B parameters per token. Text, image and audio in; text out. Strong on agentic and long-horizon work.

Grok 4.6

x-ai/grok-4.6

·xAI

xAI's frontier model, with strong coding, knowledge-work and STEM performance and a 500k-token context. Accepts text and images. Served through a third party; weights are not published.

Muse Glimmer 30B

meta-models/muse-glimmer-30b

·Meta

Muse Glimmer is Meta's 30B dense vision-language model tuned for agent workloads: tool use, long-horizon tasks and failure recovery. It accepts text and image input, supports a 131K context window, tool calling and structured output, and carries a January 2026 knowledge cutoff. Built to run always-on agents; strong on settlement-style arithmetic and reliable tool selection.

Qwen3.6 35B A3B

qwen/qwen3.6-35b-a3b

·Qwen

Qwen3.6 35B-A3B is a mixture-of-experts model with 35.9B total parameters and roughly 3B active per token, giving the quality of a large model at the speed of a much smaller one. It accepts text and image input, supports a 262K context window, and handles tool calling and structured output. Well suited to coding agents and long-context work where a whole repository or document set has to stay in the conversation.

Qwen3.8 27B

qwen/qwen3.8-27b

·Qwen

Qwen3.8 27B is a dense hybrid model — 48 Gated DeltaNet linear-attention layers and 16 gated full-attention layers — with native text, image and video input and a 262K context. Thinking is on by default with tunable reasoning effort, and it posts the family's strongest agentic scores to date (OSWorld-Verified 84.3, AndroidWorld 81.9). Built for computer-use and long-horizon agent work on a single GPU.

Qwen3 8B

qwen/qwen3-8b

·Qwen

Qwen3 8B is a dense instruction-tuned model with a hybrid thinking mode for switching between reasoning and direct answers. It offers strong general reasoning and multilingual performance for its size, with tool calling and a 32K context window. A sensible default for agent loops and high-volume tasks that do not need a frontier model.

MiniMax-M3 is a multimodal foundation model from MiniMax. It supports text, image, and video inputs with text output, a 1M-token context window, and is suited for long-horizon agentic work, coding,...

GLM-5.3-Flash is a native multimodal model from Z.ai. It is suited for efficient coding and long-horizon agent tasks. Its hybrid sparse and linear attention architecture maintains accurate long-context behavior while...

GLM-5.3 is a large-scale reasoning model from Z.ai, built for complex software engineering and long-horizon agent tasks. It supports text input and output with a 1M-token context window, and improves...

GLM 5.2 is a large-scale reasoning model from Z.ai. It supports text input and output with a 1M-token context window, and is suited for long-horizon agent workflows, project-level software engineering,...

Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort:...

MiMo-V2.5 is a native omnimodal model by Xiaomi. It delivers Pro-level agentic performance at roughly half the inference cost, while surpassing MiMo-V2-Omni in multimodal perception across image and video understanding...

Ornith 1.5 35B A3BCurrently unavailable

ornith-ai/ornith-1.5-35b-a3b

·Ornith

Currently unavailable for maintenance.

A 35B mixture-of-experts model activating roughly 3B parameters per token, with a 262,144-token context. Built for coding and long-horizon agentic work, and strong on SWE-Bench and terminal benchmarks for its size.

Ornith 1.5 9BCurrently unavailable

ornith-ai/ornith-1.5-9b

·Ornith

Currently unavailable for maintenance.

A dense 9B model with a 262,144-token context, the lightweight member of the Ornith 1.5 family. Built for coding and agentic work at a size that runs on a single GPU.

Ornith 1.5 397BCurrently unavailable

ornith-ai/ornith-1.5-397b

·Ornith

Currently unavailable for maintenance.

The flagship of the Ornith 1.5 family: a 397B mixture-of-experts model activating ten experts per token, with a 262,144-token context. Built for agentic coding and long-horizon software work, scoring 86.1 on Terminal-Bench 2.1 and 86.0 on SWE-bench Verified.