Skip to main content

Fireworks AI Models

323 entities

Kimi K3.1 Pro DSpark (RedHatAI)

model

Imported from https://huggingface.co/RedHatAI/Kimi-K3.1-Pro-DSpark at revision a23f4cf46647ebd00f5db8c03d2129c2d7030cc4

Kimi K3.1 Pro DSpark (RedHatAI)

model

Imported from https://huggingface.co/RedHatAI/Kimi-K3.1-Pro-DSpark at revision a23f4cf46647ebd00f5db8c03d2129c2d7030cc4; config.json adds top-level vocab_size/hidden_size copied from transformer_layer_config

Beam-1.0-501B-A23B-NVFP4

model

Imported from https://huggingface.co/reflection/Beam-1.0-501B-A23B-NVFP4 at revision f8ecac0e16045babf60156f6ba996a77fcb48cf9 (native NVFP4, FireTitan training base)

Ripple-65B Preview

model

Ripple-65B Preview is an open-weight, 65B-parameter model from Ricursive for general-purpose, coding, and agentic workloads. It is a general-purpose variant of the proprietary model powering the Ricursive AI chip design platform, pretrained on 12 trillion tokens. Ricursive reports a score of 80.52 on TerminalBench 2.1. Context length: up to 1M tokens. Input: text. Output: text. License: Apache 2.0.

Kimi K3.1 Pro

model

Kimi K3.1 Pro is an open-weight, natively multimodal agentic model from Moonshot AI. Post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of downstream domains, it is the first model in the K3.1 series. With a 1-million-token context window, it delivers substantial improvements in coding, 3D creation, financial research, factual accuracy, and other agentic capabilities. Its Mixture-of-Experts architecture has 2.8T total parameters, 104B activated parameters and 93 layers. It supports text and image inputs and uses MXFP4 weights with MXFP8 activations through quantization-aware training.

Ultramarine 1 Flash

model

Ultramarine 1 Flash is a multimodal model built for knowledge work in enterprise operations, finance, law, and consulting. It’s based on GLM-5.3 Flash, with 320B total parameters (18B active) and a 1M-token context window. This model is hosted in the global region. If you require region restrictions, please contact your account team. Serverless may run the model at a different numerical precision than was used in training, so outputs can differ slightly from the dedicated deployments and published evaluation results.

MiniMax M3.1 Flash

model

MiniMax-M3.1-Flash is a native multimodal model with 1M context running ~428B parameters and ~23B activated parameters. Built for agentic reasoning, tool use, coding, and long-context tasks, it keeps reasoning always on with adjustable effort from low to max. M3.1 Flash ranks among the top performers on long-horizon agentic benchmarks, scoring 62.8 on AutomationBench-AA and 73.7% on DeepSWE v1.1.

Kalshi-tuned Muse Glimmer 30B

model

Frozen Kalshi decision-study LoRA, r64/alpha32, with fitted per-question-type temperatures. Research checkpoint; no Polymarket training or validation yet.

Rune 26B-A4B

model

Rune 26B-A4B is a System One decision model: it does not reason and does not generate text. It reads a document once and answers up to 256 typed questions about it - multiple choice, true/false probability, or graded scale - returning calibrated probabilities in that single pass. Output tokens are always zero, so you pay for input only. 262K context, text only.

DeepSeek V4.1 Flash 1017

model

DeepSeek-V4.1-Flash-1017 is a minor update to DeepSeek-V4.1-Flash, delivering substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks. Alongside these gains, the model reduces task execution time and output length, enabling users to complete tasks faster and at lower cost and advancing the performance–cost Pareto frontier across multiple agent harnesses.

Step 5 Preview

model

Step 5 Preview is an open-weight model from StepFun for coding, software engineering, and agentic workloads. It uses a sparse Mixture-of-Experts architecture with 600B total / 27B active parameters across 92 layers, with Sparse GQA attention for long context. Context length: 128K tokens. Input: text. Output: text.

Reflection Beam-1.0-501B-A23B

model

Reflection Beam 1.0 is an open-weight model from Reflection AI for coding, reasoning, and agentic workloads. It uses a sparse Mixture-of-Experts architecture with 501B total / 23B active parameters, pretrained on 23.8T tokens and post-trained with reinforcement learning on 100M+ rollouts.

Muse Spark Open 816B-A42B NVFP4

model

Muse Spark Open 816B-A42B NVFP4 checkpoint (meta-models/Muse-Spark-Open-816B-A42B-NVFP4), ModelOpt NVFP4 experts (excl. layers 0/61, attn, lm_head), BF16 MoE, 1M context, multimodal (image/video/audio).

NVIDIA Nemotron 3.5 Super VL 120B A12B BF16

model

NVIDIA Nemotron 3.5 Super VL (omni: text + image), BF16. NGC final GA checkpoint ea20261008 / bf16-ga-20261003.

Muse Spark Open

model

Muse Spark Open is an open weights reasoning model from Meta for coding and agentic workflows. It supports tool calling and native image understanding in a single checkpoint. It uses a sparse Mixture-of-Experts architecture with 816B total / 42B active parameters, routing each token to 8 of 256 experts, and hybrid attention that repeats three sliding-window layers (2,048-token window) for every full-attention layer. Inputs: text and images. Output: text.

NVIDIA Nemotron 3.5 Super VL 120B A12B NVFP4

model

NVIDIA Nemotron 3.5 Super VL is an open multimodal model built for long-running agents. It combines reasoning, instruction following, and tool use with text, image, document, and video understanding in a single model. It uses a hybrid Mamba-Transformer Mixture-of-Experts architecture with 120B total / 12B active parameters, plus Multi-Token Prediction (MTP) for faster generation. Inputs: text, images, documents, and video. Output: text.

Drex v2.0

model

Drex v2.0 decision model

MiMo-V2.6-Pro-RL

model

MiMo-V2.6-Pro-RL is a 1T-parameter sparse MoE omnimodal model (text, image, video, audio) with 1M-token context, designed as Xiaomi’s flagship agentic model. It scales a unified reinforcement-learning pipeline across coding, general agents, visual tasks, and cybersecurity using groupwise agentic grading and speculative decoding to drive self-improvement and strong benchmark performance

Ember-1

model

Ember-1 is a specialized model from Fireworks. Built on Kimi K3, it produces shorter reasoning traces, using approximately 40% fewer tokens while maintaining comparable quality across our evaluations

DeepSeek V4.1 Flash

model

DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with 552B backbone parameters that natively processes images and text at up to 1M token context. Its Causal Encoder-Decoder architecture activates only 8B parameters during prefill and 16B during decode, cutting the KV cache footprint to roughly a quarter of DeepSeek-V4-Flash for cost-efficient agentic workloads.

Ling 3 Flash Fin

model

Ling-3.0-flash-Fin is the first finance-enhanced model in the Ant Ling family. Developed by Ant Group with leading financial institutions and domain experts, it extends Ling-3.0-flash through continued training on high-quality financial data. With 124B total parameters, 5.1B activated parameters, and a 256K context window, the model combines financial expertise with efficient inference for long-horizon agent workflows.

DeepSeek-V4-Flash-Vision-Exp

model

DeepSeek-V4-Flash-Vision-Exp is the first experimental multimodal model in the DeepSeek-V4 family. It builds on DeepSeek-V4-Flash by adding visual modules and continued training for visual understanding, with substantially improved multimodal agent capabilities while remaining comparable on text-only agent tasks. It has the same model structure as DeepSeek-V4-Flash-DSpark, i.e. it comes with a speculative decoding module attached.

GLM 5.3 Flash

model

GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series with 320B total parameters and 18B active parameters. It incorporates several architectural improvements over GLM-5.2 including a hybrid architecture, sharply reducing long-context serving costs while preserving precise long-context and Manifold-Constrained Hyper-Connections (mHC) to further improve scaling efficiency.

Qwen3.8 Flash Next FP8

model

Qwen3.8-Flash-Next is an experimental Qwen4-architecture preview: multimodal MoE, 125B parameters (6B active) plus 51B n-gram embeddings, native 262K context.

Qwen3.8 Flash Next FP4

model

Qwen3.8-Flash-Next is an experimental Qwen4-architecture preview: a multimodal MoE with 125B parameters (6B active) plus 51B n-gram embeddings and native 262K context.

Page 1 of 13