AI Directory · Researched reviews · ASEAN lens
LLMs & Chat
134 tools · LLMs & Chat · Most popular
How to choose a language model
Capability is the wrong first question. The frontier models sit close enough on general reasoning that the gap rarely decides anything, and the ranking changes every few weeks anyway. Three cheaper checks decide it: what you pay at your real volume, how much context you need in one call, and whether you can send the data at all.
Compare price per million tokens, not the monthly subscription. Look at input and output separately; output typically costs three to five times more, and most workloads are output-heavy. A model that looks twice as expensive per token can be cheaper per finished task if it needs fewer retries.
Context length matters less than people assume, right up to the moment it is the only thing that matters. Summarising a contract, reading a codebase, or holding a long conversation are the cases where a small window forces you into chunking, and chunking is where accuracy goes.
For teams in Singapore, Malaysia and Indonesia, the constraint is often neither price nor context; it is where inference runs. If your data cannot leave the jurisdiction, most of the shortlist disappears before capability is discussed. Open-weight models you host yourself become the practical answer, not the ideological one.
Also listed in LLMs & Chat
44 more tools we track but have not written up in depth. Each entry links to the vendor's own site; we have not tested these tools.
-
360 Zhinao 360智脑 Qihoo 360's model family: open 7B weights and a hybrid per-token API
-
Amazon Nova Amazon's low-cost foundation models on Bedrock, led by Nova 2 Lite
-
Awarri Lagos company behind N-ATLAS, Nigeria's capped-licence language model
-
Baichuan 百川智能 Wang Xiaochuan's lab, now focused entirely on medical AI
-
Baseten Production AI inference on autoscaling GPUs, with Model APIs and training
-
Beluga Stability AI's 2023 Llama fine-tunes — non-commercial, abandoned
-
Chai AI Mass-market AI roleplay app with millions of user-made characters
-
Comet by Perplexity Perplexity's free AI browser with an assistant that acts on pages
-
Dia by The Browser Company AI browser for work from the makers of Arc
-
Dolly Databricks' 2023 instruction model, remembered for its human-written dataset
-
Falcon (TII) Abu Dhabi's hybrid Mamba-Transformer open-weight models
-
Hanooman India's Indic LLM family (1.5B–40B) — consumer app pulled in 2025
-
Hyperbolic GPU cloud for on-demand and reserved H100 to B300 instances
-
Jais (Inception42 / G42) Open-weight Arabic-English LLMs from Abu Dhabi, now Jais 2 at 70B
-
KARAKURI (カラクリ) Tokyo customer-support AI agents, built on its own Japanese models
-
Kindroid Power-user AI companion: deep memory, voice calls, and unfiltered roleplay
-
Koala Berkeley's 2023 dialogue-tuned LLaMA — a research-era milestone
-
KT Mi:dm KT's sovereign Korean-first LLM — MIT-licensed, enterprise-ready
-
Lambda Inference Pay-per-token inference API that Lambda itself is winding down
-
Latam-GPT CENIA's open-weight 70B model tuned on Latin American Spanish and Portuguese
-
Liquid AI Efficient hybrid AI models built for on-device inference
-
Llama Guard Meta's open-weight classifier for unsafe LLM prompts and replies
-
Mixtral Mistral's open-weight mixture-of-experts model, Apache 2.0 licensed
-
Nomi AI AI companion with three-tier memory, voice calls, and a public API
-
NTT tsuzumi NTT's from-scratch Japanese LLM sized to run on a single GPU
-
NVIDIA Nemotron NVIDIA's open-weight model family built for agentic workloads
-
Olmo by Ai2 Fully open LLMs: weights, data, code and checkpoints
-
OPT Meta's 2022 research LLMs, released with code and training logbooks
-
Phi Microsoft's small, MIT-licensed language models for local and low-cost use
-
Platypus 2023 leaderboard-topper built from a 25k-example STEM dataset and a model merge
-
Poe by Quora Quora's app for chatting with many AI models and bots on one plan
-
RedPajama Open LLaMA-recipe training data — 30T tokens, fully transparent
-
Requesty AI gateway routing 600+ models with EU residency and a 5% markup
-
rinna Japan's open-weight AI catalog — commercial API discontinued in 2026
-
Sakana AI Tokyo AI lab building nature-inspired foundation models
-
SambaNova Cloud Fast open-model inference on SambaNova's own RDU chips
-
Shisa.ai Tokyo lab with open Japanese LLMs and pay-as-you-go speech APIs
-
Sider AI A multi-model AI sidebar and browser agent for Chrome, Edge and Safari
-
Skywork 天工 Kunlun Tech's AI agent platform for research, documents, slides and spreadsheets
-
Smaug by Abacus Abacus.AI's 2024 open-weight fine-tunes, now legacy
-
SpicyChat Uncensored AI roleplay with 300,000+ user-made characters
-
StableLM Stability AI's dormant open LLM family — no new models since 2024
-
Tencent HY (Hunyuan) 腾讯混元 Tencent's HY model family, from open-weight LLMs to video and 3D
-
Tülu 3 (Ai2) Ai2's open recipe for turning base LLMs into instruct models