/aicg/ meta
Comparison between the different services/models and frontends used by /aicg/.
These ratings aren't gospel. They're the opinion of one anon who tried to incorporate suggestions and push no agenda.
Services/models
If you're curious about a service, the OP should have the information you need.
🏆 - Best
🥈 - Good
🥉 - Usable
🌱 - Bad
⛔ - Worst
🔀 - Depends
NSFW - general ERP-readiness (taking into account the need to jailbreak)
SFW - general RP-readiness
MEMORY - more size than retention, more memory = higher price for both cloud and local
CREATIVITY - low only for dumb assistants, can be increased with a CoT prompt
COMPLEX INSTRUCTIONS - stats tracking, logical conditions, multiple characters, etc.
PRICE - per token, per month, or for running local models one time purchase or per hour rental of the hardware
EASE OF USE - OAI keys or all proxies are easy, but still require JBs; Horde and subscription based local/NAI offer inbuilt nobrain presets; running local is not that hard because there's enough guides; anything can require shuffling prompt parts and finding the best temperature and other settings
/lmg/ - local models in general, the ones you can download and run offline, the best ones in the rankings
| NSFW | SFW | MEMORY | CREATIVITY | COMPLEX INST | PRICE | EASE OF USE | |
|---|---|---|---|---|---|---|---|
| GPT-4.5 | 🥈 | 🏆 | 🏆(128k) | 🏆 | 🏆 | HIGHEST | 🥈 |
| GPT-4.1 | 🏆 | 🏆 | 🏆(1kk) | 🏆 | 🏆 | MEDIUM | 🥈 |
| GPT-4 | 🏆 | 🏆 | 🥈(8k/32k) | 🏆 | 🏆 | HIGH | 🥈 |
| GPT-4 Turbo | 🥈 | 🥈 | 🏆(128k) | 🥈 | 🏆 | MEDIUM | 🥈 |
| GPT-4o | 🥈 | 🥈 | 🏆(128k) | 🥈 | 🏆 | MEDIUM | 🥈 |
| Claude 4.0 Opus | 🏆 | 🏆 | 🏆(200k) | 🏆 | 🏆 | HIGH | 🥈 |
| Claude 4.0 Sonnet | 🏆 | 🏆 | 🏆(200k) | 🥈 | 🏆 | MEDIUM | 🥈 |
| Claude 3.5 Haiku | 🥉 | 🥉 | 🏆(200k) | 🥉 | 🌱 | LOW | 🥈 |
| Gemini 2.5 Pro | 🥈 | 🥈 | 🏆(2kk) | 🥈 | 🏆 | FREE/MEDIUM | 🥈 |
| Gemini 2.5 Flash | 🥈 | 🥈 | 🏆(2kk) | 🥈 | 🥉 | FREE/LOWEST | 🥈 |
| Grok | 🥈 | 🥈 | 🏆(131k) | 🥈 | 🥉 | MEDIUM | 🥈 |
| DeepSeek-V3-0324 | 🏆 | 🏆 | 🔀 | 🥈 | 🥈 | LOWEST | 🥈 |
| DeepSeek-R1 | 🏆 | 🏆 | 🔀 | 🏆 | 🏆 | LOW | 🥉 |
| Mistral Large 2.1 | 🥈 | 🥈 | 🏆(128k) | 🥈 | 🥈 | FREE/LOW | 🥈 |
| Command A | 🥈 | 🥈 | 🏆(256k) | 🥈 | 🥈 | FREE | 🥈 |
| Jamba 1.6 Large | 🥈 | 🥈 | 🏆(256k) | 🥉 | 🥉 | FREE/MEDIUM | 🥈 |
| NovelAI (Erato) | 🥈 | 🥈 | 🥉(8k) | 🥈 | 🥉 | MEDIUM | 🌱 |
| NovelAI (Kayra) | 🥈 | 🥈 | 🥉(8k) | 🏆 | 🌱 | MEDIUM | 🌱 |
| NovelAI (Clio) | 🥉 | 🥉 | 🥉(8k) | 🥈 | ⛔ | LOW | 🌱 |
| CAI | 🌱 | 🥈 | ⛔(2k/3k) | 🥈 | ⛔ | FREE | 🏆 |
| /lmg/ 7B+ | 🥉 | 🥉 | 🔀 | 🥉 | 🌱 | LOW | 🥉 |
| /lmg/ 12B+ | 🥈 | 🥈 | 🔀 | 🥈 | 🌱 | LOW | 🥉 |
| /lmg/ 70B+ | 🥈 | 🥈 | 🔀 | 🥈 | 🥉 | HIGH | 🥉 |
| /lmg/ 100B+ | 🥈 | 🥈 | 🔀 | 🥈 | 🥈 | HIGH | 🥉 |
| /lmg/ Horde/proxy | 🥈 | 🥈 | 🔀 | 🥈 | 🔀 | FREE | 🥈 |
| /lmg/ subscription | 🥈 | 🥈 | 🔀 | 🥈 | 🔀 | LOW/MEDIUM | 🥈 |
Notes:
- Gemini API: free access is rate limited.
- Cohere API: free access is rate limited.
- AI21 Jamba API: free trial is limited.
- NovelAI models:
- Not intended for chat usage and not recommended.
- But since they're bundled with Image Gen sub, you might want to experiment.
- Local models:
- Check out /lmg/
- Small brain/short memory models can be run on very basic hardware, but size+memory+speed vs. requirements curve is very steep.
- Settings and presets are shared as a single file for easy import and setup.
- OpenAI o1/o3: thinking models by OpenAI aren't worth using for RP
- DeepSeek-R1: enjoyable but at the same time has annoying issues.
Quick notes on how to access each:
- GPT:
- Official API, revendors.
- Or use a reverse proxy that has it. (agnai video guides / tavern guide)
- Claude: Official API, revendors.
- Claude (AWS): Pay for it https://aws.amazon.com/bedrock/pricing/
- Gemini API: Official API through AI Studio (https://makersuite.google.com/app/apikey).
- Mistral: Official API (https://console.mistral.ai).
- Cohere: Official API (https://dashboard.cohere.com).
- Grok: Official API (https://console.x.ai/).
- DeepSeek: Official API (https://platform.deepseek.com).
- Jamba: Official API (https://studio.ai21.com).
- /lmg/
- Use a reverse proxy or Horde.
- Pay for a subscription that offers API access. Featherless (https://featherless.ai/#pricing) might be a good choice.
- Some services offer free limited API access to small models.
- Or learn to run it locally, on Colab or via GPU rental services.
Most models are available on OpenRouter (https://openrouter.ai/models) or from other revendors.
Frontend (service support) (outdated)
- Character Card V3 Implementation Tracking: https://rentry.org/zdwfrgqq
Three frontends are currently recommended, Silly, Agnai, and Risu. All have a lot of different features and are improving fast.
Miku is noteworthy, but difficult to compare. It uses sentiment classification to show different sprites for each message, and text-to-speech, supporting OAI, Kobold, Whisper, AzureTTS, ElevenLabs, and NovelAITTS.
| Agnai | Risu | Silly | Miku.gg | |
|---|---|---|---|---|
| OpenAI API | ✅ | ✅ | ✅ | ✅ |
| Anthropic API | ✅ | ✅ | ✅ | ⛔ |
| NovelAI API | ✅ | ✅ | ✅ | ⛔ |
| Scale | ✅ | ⛔ | ✅ | ⛔ |
| Horde | ✅ | ✅ | ✅ | ⛔ |
| Oobabooga | ✅ | ✅ | ✅ | ✅ |
| OpenRouter | ✅ | ✅ | ✅ | ⛔ |
| Kobold | ✅ | ✅ | ✅ | ⛔ |
| GooseAI API | ✅ | ⛔ | ⛔ | ⛔ |
| Replicate API | ✅ | ⛔ | ⛔ | ⛔ |
| Google API | ⛔ | ✅ | ✅ | ⛔ |
| Novelist API | ⛔ | ✅ | ⛔ | ⛔ |
| Mancer API | ✅ | ⛔ | ✅ | ⛔ |
| Gemini API | ⛔ | ✅ | ✅ | ⛔ |
| Mistral API | ✅ | ✅ | ✅ | ✅ |
Gemini works on all frontends if using it via reverse proxy. Mistral API supports OpenAI input (using the official API like a reverse proxy).
Frontend (features) (outdated)
| Agnai | Risu | Silly | Miku.gg | |
|---|---|---|---|---|
| UJB | ✅ | ✅ | ✅ | ? |
| Online access | ✅ | ✅ | ⛔ | ✅ |
| Local install | ✅ | ✅ | ✅ | ✅ |
| Multi-character chat | ✅ | ✅ | ✅ | ⛔ |
| Multi-user chat | ✅ | ⛔ | ✅ | ⛔ |
| Image caption | ⛔ | ? | ✅(ext) | ⛔ |
| Image generation | Horde, Stable Diffusion, NAI | Stable Diffusion | (ext) Horde, Stable Diffusion ) | Stable Diffusion |
| Summary | 🌱 | OAI/Claude | HF/OAI/Claude | ⛔ |
| Screenshot | ✅ | ✅ | ⛔ | ⛔ |
| TTS | Web Speech Synthesis, 11labs, NAI | Web Speech Synthesis, 11labs, VOICEVOX | 11labs, Silero, System, Edge, Coqui, NAI | 11labs, Azure, NAI |
| STT | ✅ | ⛔ | ✅(ext) | ⛔ |
| Emotion packs | ⛔ | ✅ | ✅ | ✅ |
| Scenes | ⛔ | ✅ | ⛔ | ✅ |
| Embeddings | ✅ | ✅ | ✅(ext) | ⛔ |
| Character sprites | ✅ | ⛔ | ⛔ | ⛔ |
| UI styling | ✅ | ✅ | ✅ | ✅ |
| Bias | ⛔ | ✅ | ✅ | ? |
| AI writes your message | 🌱(add a card with your name as character participant) | ✅ (auto-suggest) | ✅ (impersonate) | ? |
| Translation | ? | ✅ | ✅ | ? |
| Lorebooks | ✅ | ✅ | ✅ | ? |
| Regex search and replace | ⛔ | ✅ | ✅ | ⛔ |
| Phone use | ✅ | ✅ | 🌱 | 🌱 |
| CYOA | ✅ | ✅ (trigger feature) | ⛔ | ⛔ |
- Only agnai has support for choosing which character speaks at any time in multi-user rooms, although on the flipside it doesn't support fully automatic mode
- (ext) means you have to install the Silly extras. They can be a little complicated to setup for non-technical users.
Outdated uninteresting information
| NSFW | SFW | MEMORY | CREATIVITY | COMPLEX DEFS | PRICE | EASE OF USE | |
|---|---|---|---|---|---|---|---|
| Claude (Poe) | ⛔ | 🥈 | 🌱 | 🏆 | 🌱 | FREE | 🏆 |
| Turbo (Poe) | 🥈 | 🥈 | 🌱 | 🥈 | 🌱 | FREE | 🏆 |
| Claude (Slaude) | 🥈 | 🏆 | 🌱 | 🏆 | 🥈 | FREE | 🌱 |
| Pyg (LLaMA) | 🥈 | 🥈 | 🌱 | 🌱 | 🌱 | FREE | 🌱 |
| NovelAI (old) | 🌱 | 🌱 | 🌱 | 🌱 | 🌱 | MEDIUM | 🥈 |
| Pyg 6B | 🌱 | ⛔ | 🌱 | 🌱 | 🌱 | FREE | 🥈 |
| Davinci-3 | 🥈 | 🥈 | 🌱(4k) | 🥈 | 🌱 | MEDIUM | 🏆 |
| Claude (API) | 🏆 | 🏆 | 🏆(100k) | 🏆 | 🥈 | FREE | 🥈 |
| Claude Instant (AWS) | 🥈 | 🥈 | 🏆(18k/100k) | 🥉 | 🌱 | LOW | 🌱 |
| Claude (AWS) | 🏆 | 🏆 | 🏆(200k) | 🏆 | 🏆 | HIGH | 🌱 |
| /lmg/ Mixtral | 🥈 | 🥈 | 🥈(4k-32k) | 🌱 | 🥈 | LOW | 🥉 |
| Mistral Medium | 🥈 | 🥈 | 🥈(32k) | 🥉 | 🥉 | LOW | 🥈 |
| Gemini Advanced | 🥉 | 🏆 | 🏆(32k) | 🏆 | 🥈 | MEDIUM | 🥉 |
| Claude 3.0 Sonnet | 🏆 | 🏆 | 🏆(200k) | 🥈 | 🥈 | MEDIUM | 🥈 |
| Claude 3.0 Haiku | 🥉 | 🥉 | 🏆(200k) | 🥉 | 🌱 | LOWEST | 🥈 |
| GPT-3.5 Turbo | 🌱 | 🌱 | 🥈(4k/16k) | 🌱 | 🌱 | LOWEST | 🥈 |
| Gemini 1.5 Flash | 🥉 | 🥉 | 🏆(2kk) | 🥉 | 🌱 | FREE/LOWEST | 🥈 |
| Gemini 1.5 Pro | 🥈 | 🥈 | 🏆(2kk) | 🥈 | 🥉 | FREE/MEDIUM | 🥈 |
| o1 | 🥉 | 🥉 | 🏆(200k) | 🌱 | 🏆 | HIGH | 🥈 |
| o3-mini | 🌱 | 🌱 | 🏆(200k) | 🌱 | 🥈 | LOW | 🥈 |
| Claude 3.5 Sonnet | 🏆 | 🏆 | 🏆(200k) | 🥈 | 🏆 | MEDIUM | 🥈 |
| Command R+ | 🥈 | 🥈 | 🏆(128k) | 🥈 | 🥈 | FREE | 🥈 |
| Claude 3.0 Opus | 🏆 | 🏆 | 🏆(200k) | 🏆 | 🏆 | HIGH | 🥈 |
| Claude 3.7 Sonnet | 🏆 | 🏆 | 🏆(200k) | 🥈 | 🏆 | MEDIUM | 🥈 |
| Gemini 2.0 Pro | 🥈 | 🥈 | 🏆(2kk) | 🥈 | 🥈 | FREE/MEDIUM | 🥈 |
| Gemini 2.0 Flash | 🥈 | 🥈 | 🏆(2kk) | 🥈 | 🥉 | FREE/LOWEST | 🥈 |
| DeepSeek-V3 | 🥈 | 🥉 | 🔀 | 🥉 | 🥉 | LOWEST | 🥉 |
| GPT-4o mini | 🌱 | 🌱 | 🏆(128k) | 🌱 | 🥉 | LOWEST | 🥈 |
- Poe: Create a bot over at https://poe.com/. Example: https://poe.com/coolreisen
- Slaude: it's still viable for sfw purposes for anons with actual enterprise accounts
- Bing: there are ways to use this version of GPT-4 for free, but it's "filtered"
- Clewd: there are ways to use claude.ai version of Claude for free, but it's "filtered"
- text-davinci-003 has been removed as OpenAI is discontinuing it.
- Davinci-3: Use an OpenAI key directly in SillyTavern/agnai. You can use your own or buy one (search "buy openai key" on yandex).
- Turbo: Same as davinci-3 but you can also use a reverse proxy.
- Gemini Advanced: the rating is provided for Gemini 1.0 Ultra, which was served to subscribers in the beginning. Pay for a subscription.
- Consensus on Gemini 1.5 Pro is: better than 3.5 Turbo, worse than 4 Turbo.
- Grok API: free credits until the end of 2024.
- DeepSeek-V3: usable but at the same time has severe issues.