Signal & Noise
Menu

Model Tracker

The models that matter right now, in one table — curated editorially, so it reflects what's actually worth using rather than everything that exists.

Prices are per 1M tokens. Last reviewed 2026-07-15— “verify” means check the provider's current pricing page before budgeting.

ModelProviderContextInputOutputWeightsBest for
Claude Fable 5Anthropic1M$10.00$50.00closedHardest reasoning and long-horizon agentic work; always-on thinking
Claude Opus 4.8Anthropic1M$5.00$25.00closedFlagship coding and agentic work; long autonomous runs
Claude Sonnet 5Anthropic1M$3.00 ($2 intro)$15.00 ($10 intro)closedNear-Opus coding quality at Sonnet cost; the volume workhorse
Claude Haiku 4.5Anthropic200K$1.00$5.00closedFast, cheap classification, extraction, and routing
GPT-5.x (latest)OpenAI400K+verifyverifyclosedGeneral-purpose flagship; broad ecosystem and tooling
o-series (reasoning)OpenAI200K+verifyverifyclosedHard math, science, and multi-step reasoning at higher latency
Gemini 3 ProGoogle1M+verifyverifyclosedVery long context, multimodal breadth, Workspace integration
Llama 4 familyMetaup to 1M+self-hostself-hostopenOpen-weights general models; fine-tuning base for enterprises
DeepSeek V3.x / R1DeepSeek128Kvery lowvery lowopenOpen-weights reasoning at aggressive cost; MoE efficiency
Qwen 3 familyAlibaba128K+self-hostself-hostopenStrong multilingual open weights across many sizes; agent tuning
Mistral (Large / small)Mistral128KverifyverifyopenEuropean provider; efficient open and hosted models
Gemma 3 familyGoogle128Kself-hostself-hostopenSmall open models that run on a single GPU or laptop