Skip to content
NerdLevelTech

Leaderboard

Model leaderboard

Picks by use case first. Then the numbers behind them, every one of them cited.

Last verified: October 24, 2026Model indexHow we rank

Before you read the table

Everything in this table is sourced: prices, context windows and release dates come from the maker’s own pages, and each model page links the exact source and the date it was checked. We do not run a benchmark harness, so we do not publish a score — the picks are editorial, and each one names the fact behind it.

Our picks

Editorial, not a benchmark. Each pick names the fact that earned it, and every model page shows the caveat.

  • Best overall

    Claude Opus 5.5

    Anthropic recommends it as the starting point for most workloads, and it backs that up with a 1M context, documented tool use and a $4/$20 price. The default we would ship first.

    NLT pickPaid API$4.00 / $20.00

  • Best value

    GPT-6.1 Sol

    OpenAI positions it as near-Astra performance for complex coding at $2/$10, with the same 1.05M context. The best quality-per-dollar on this list.

    NLT pickPaid API$2.00 / $10.00

  • Best budget workhorse

    DeepSeek V4.1 Flash

    The cheap agent pick: vision, a 1M-token context window and tool use at about $0.027 per task, half that off-peak. GLM-5.3 Flash is cheaper for text-only work; this is the one we would hand an agent.

    NLT pickOpen weights$0.30 / $1.20

  • Best open weights

    GLM-5.3 Flash

    The open-weight pick: a dedicated Hugging Face repo, an Ollama library, a 1M context and native image input at $0.15/$0.50 per 1M. The full GLM-5.3 costs about nine times more for the same window.

    NLT pickSelf-host or API$0.15 / $0.50

The numbers

Sort any column. Prices are USD per 1M tokens on the base tier, read from the vendor page on the verification date.

Models are grouped by maker. Expand a maker to compare.

Model

Task cost = 50K input + 10K output tokens. Caching, batch discounts and tier crossings are ignored on purpose so the column stays comparable.

How we rank models

Three kinds of claims, kept apart on purpose: verified facts, computed numbers, and editorial picks.

Read the methodology