Aurekai

Model Commons

Models you can try

Pick a model to see what it does, try it on your own text, then use it in your work.

Models
25
Things they do
184
Connections
175
Broken links
547
Qwen2.5 Coder 14B Instruct (native FPQ pack)Native Bonfyre FPQ sharded pack manifest for Qwen2.5-Coder-14B-Instruct.native-fpq0-sharded · 14.8 GBNot used anywhere yetdoes 0 · used by 0 · 0 made from itLlama 3 70B Instruct (FPQ compressed)bonfyre-fpq INT8 compressed Llama 3 70B Instruct. ~9.2 GB. A3 --tier pro inference stage.bin · 9.2 GBNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenLlama 3 8B InstructMeta Llama 3 8B Instruct GGUF Q4_K_M. 4.7 GB. Brief extraction, repurpose, and summary stages.gguf · 4.7 GBNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenWhisper Large v3OpenAI Whisper large-v3. 99-language ASR, 2.9 GB GGUF. Used by all A/M/V/R series recipes.gguf · 2.9 GBUsed in Podcast Multi-Modal Repurposingdoes 0 · used by 1 · 0 made from it · some links brokenLlama 3 8B Instruct (FPQ compressed)bonfyre-fpq INT8 compressed Llama 3 8B Instruct. ~1.2 GB. Drop-in replacement; load with BonfyreFPQ proxy.bin · 1.2 GBNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenWhisper MediumOpenAI Whisper medium. 769 MB GGUF. Balanced accuracy/speed; A2 archive default.gguf · 769 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenOuteTTS 0.3 500M (GGUF Q4_K_M)Neural text-to-speech language model from OuteAI, run natively by llama.cpp's llama-tts with the WavTokenizer vocoder. Licence CC BY-SA 4.0 (model card, 2026-11-08); audio it renders carries that attribution. It powers capability narrate; registration implies no measured voice quality and no right to imitate a real speaker.gguf · 385 MBUsed in Produced Audio Programsdoes 1 · used by 1 · 0 made from itBitNet Embedding 270M I2_SOfficial BitNet embedding release body used only through the distinct BitNet representation passport. This identity does not imply comparability with any Bonfyre or provider embedding geometry.gguf · 350 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from itBGE Reranker BaseBAAI BGE reranker base. 278 MB GGUF. FPQ scoring stage in A3 full pipeline.gguf · 278 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenWhisper BaseOpenAI Whisper base model. 148 MB GGUF. Fast transcription for quick briefs (A1 default on low-memory systems).gguf · 148 MBUsed in Audio Quick Brief and 2 moredoes 0 · used by 3 · 0 made from it · some links brokenWavTokenizer Large 75 (GGUF F16)Audio codec decoder that turns OuteTTS audio tokens into 24 kHz waveform under llama-tts. Upstream novateur/WavTokenizer-large-speech-75token, MIT; GGUF conversion by ggml-org. A vocoder alone speaks nothing.gguf · 124 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from itNomic Embed Text v1.5Nomic AI embedding model. 274 MB GGUF. Used by BonfyreEmbed for vector indexing in A2/A3/R1.gguf · 80 MBUsed in Embeddingdoes 1 · used by 1 · 0 made from it · some links brokennomic-embed-text v1.5 Q2_K localLive HF artifact staged locally for BonfyreModel publish validation.gguf · 49 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from itPyannote Speaker Segmentation 3.1Speaker diarisation model. 17 MB ONNX. BonfyreSegment stage. Required for A3/M1 multi-speaker.onnx · 17 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenNomic static embedding, 128 dimensionsStatic token embedding distilled by bonfyre-embed distill from the local nomic-embed-text-onnx graph (PCA to 128 dimensions, Zipf weighting, int8 rows). Static semantic search reads it in CMS publishing, the web address server and the browser. This identity is not the teacher's sentence model and implies no retrieval quality; what has been measured is carried by its representation passport.bfse · 4 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from itSilero VADVoice activity detection. 1.8 MB ONNX. BonfyreIngest stage gates silence removal before transcription.onnx · 2 MBUsed in Full Speech Investigation Pipelinedoes 0 · used by 1 · 0 made from it · some links brokenQwen2.5 Coder 14B Instruct TokenizerTokenizer JSON for the native Qwen2.5-Coder-14B-Instruct FPQ pack.tokenizer-json · 1 MBNot used anywhere yetdoes 0 · used by 0 · 0 made from itT04 Topic Mapper (Global)Collapsed student MLP for topic classification. Global geometry — works across short/long docs. Mean F1 0.914, 10/12 pass, stability 0.895. Trained via T04 recipe: SBERT embed → MLP student → collapse.onnx · under 1 MB · testedNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenT15 Topic Mapper (Global, High-Stability)Collapsed student MLP, augmented contrastive training. Global geometry. Mean F1 0.911, 12/12 pass, stability 0.876. Highest pass rate of any family.onnx · under 1 MB · testedNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenT16 Topic Mapper (Long-Form Conditional)Collapsed student MLP with document-length conditioning. Best on long-form corpora (avg_doc_len > 500). Mean F1 0.931 conditional, 5/8 pass. Geometry condition: avg_doc_len > 500.onnx · under 1 MB · testedNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenMistral EmbedRemote embedding-model identity. Its provider vectors are recorded separately as typed latent representations; this catalog row is not a local model body or a claim of retrieval quality.provider-api · size not recordedNot used anywhere yetdoes 0 · used by 0 · 0 made from itNomic Embed Text v1.5 (ONNX int8)768-dim sentence embedder, int8 ONNX export. Backs bonfyre-embed --backend onnx (onnx-runtime-native). Layout: <cache>/nomic-embed-text-onnx/{onnx/model.onnx,vocab.txt}.onnx · size not recordedNot used anywhere yetdoes 0 · used by 0 · 0 made from itQwen2.5 3B Instruct (GGUF Q4_K_M)Local generation model. Sideloaded because bonfyre-model pull fails non-resumably above ~1GB.gguf · size not recordedNot used anywhere yetdoes 0 · used by 0 · 0 made from itTopic Mapper Collapse (T08)Compact collapsed student from cnn_dm-1000 diversity run. Small-param variant (24,770 params) using token-average pooling. F1=0.898 on cnn_dm (global geometry). Not promoted to primary routing due to lower coverage vs T04/T15.onnx · size not recorded · testedNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links brokenTopic Mapper Collapse (T14)Collapsed student from cnn_dm-1000 diversity run. Full-param variant (99,331 params). F1=0.823 on cnn_dm (global geometry). Not promoted to primary routing due to lower F1 vs T04/T15/T16.onnx · size not recorded · testedNot used anywhere yetdoes 0 · used by 0 · 0 made from it · some links broken

547 links name something that was never added to the catalogue. They are left out of these counts, and each model’s page lists its own.

What hosted models charge

US dollars per million tokens, as each provider lists it. Prices change, and a lower price says nothing about quality.

ModelReading inWriting outProviderSeen
ibm-granite/granite-4.0-h-micro$0.017—ibm-granite
openai/gpt-oss-20b$0.018—openai
mistralai/mistral-nemo$0.019—mistralai
deepseek/deepseek-v4-flash-0731$0.021—deepseek
inclusionai/ling-3.0-flash$0.021—inclusionai
inclusionai/ling-3.0-flash-vl$0.021—inclusionai
~deepseek/deepseek-v4-flash-latest$0.021—~deepseek
openai/gpt-oss-20b:batch$0.024—openai
openai/gpt-5-nano:batch$0.025—openai
meta-llama/llama-3.2-1b-instruct$0.027—meta-llama
openai/gpt-oss-120b:batch$0.030—openai
inference-net/schematron-v2-turbo$0.030—inference-net
qwen/qwen3.7-flash$0.030—qwen
amazon/nova-micro-v1$0.035—amazon
~deepseek/deepseek-flash-latest$0.035—~deepseek
cohere/command-r7b-12-2024$0.037—cohere
inception/mercury-2.5$0.040—inception
sao10k/l3-lunaris-8b$0.040—sao10k
z-ai/glm-5.3-flash$0.040—z-ai
~z-ai/glm-flash-latest$0.040—~z-ai
tencent/hy-mt2-1.8b$0.044—tencent
deepseek/deepseek-v4-flash$0.047—deepseek
google/gemini-2.5-flash-lite:batch$0.050—google
google/gemma-3-12b-it$0.050—google
google/gemma-3-4b-it$0.050—google
inference-net/schematron-v2-small$0.050—inference-net
meta-llama/llama-3.1-8b-instruct$0.050—meta-llama
meta-llama/llama-3.2-3b-instruct$0.050—meta-llama
mistralai/mistral-small-24b-instruct-2501$0.050—mistralai
nvidia/nemotron-3-nano-30b-a3b$0.050—nvidia
openai/gpt-4.1-nano:batch$0.050—openai
openai/gpt-5-nano$0.050—openai
openai/gpt-6-luna-pro:batch$0.050—openai
openai/gpt-6-luna:batch$0.050—openai
upstage/solar-mini4$0.050—upstage
amazon/nova-lite-v1$0.060—amazon
ibm-granite/granite-4.2-8b$0.060—ibm-granite
inclusionai/ling-3.0-flash-fin$0.060—inclusionai
poolside/laguna-xs-2.1$0.060—poolside
z-ai/glm-5.3-flash:batch$0.060—z-ai