Which model, right now

One answer per job — the single model to reach for today, drawn from live quality, speed, and cost data. No filters, no comparison grid.

Updated 2h ago

Hard / agentic work

Long agent loops, tool calling, and hard multi-step reasoning.

DeepSeek V4 Flash 0731

deepseekQuality 88%Context 1.3MCost $0.06/M

or try GLM 5.3 Flash·See all →

Coding

Writing and reviewing code, resolving real issues.

MiniMax M2.5

minimaxSWE-Bench 92%Cost $0.68/M

Compares API pricing only — subscription tools like Claude Code and Codex sit outside this ranking and are often the strongest option if you're already paying for a seat.

or try DeepSeek V3.2·See all →

Writing

Prose, docs, and anything a human will actually read.

Claude Opus 4.6

anthropicArena 100%Cost $15.00/M

or try Gemini 3 Flash Preview·See all →

Vision

Reading images, screenshots, and documents.

Gemini 3 Flash Preview

googleQuality 99%Cost $1.75/MContext 1.0M

or try DeepSeek V4.1 Flash·See all →

Cheap bulk tasks

High-volume tasks where cost per token dominates.

DeepSeek V4 Flash 0731

deepseekCost $0.06/MQuality 88%Speed 234 tok/s

or try Qwen3.7 Flash·See all →

Real-time / low latency

Chat UIs and voice, where time to first token matters.

DeepSeek V4.1 Flash

deepseekSpeed 237 tok/sTTFT 0.9sCost $0.38/M

or try Claude Haiku 4.5·See all →

Huge documents

Whole codebases and huge documents in one prompt.

DeepSeek V4 Flash 0731

deepseekContext 1.3MCost $0.06/MQuality 88%

or try GLM 5.3 Flash