LLM·Dex
Newsletter
Use caseTop 6 picks

Cheapest LLMs in 2026

Lowest dollar-per-million-tokens models that still meet quality bars.

Updated

How we ranked

  • Output price per 1M tokens
  • Quality floor, must clear basic instruction-following
  • Latency
  • Context window adequacy (≥32k)
  • API stability and rate-limit headroom

Read the full methodology for our sourcing and ranking standards.

Price-per-token has fallen roughly 90% since 2023, and the cheapest models in 2026 are now genuinely good, not the lobotomized cousins they used to be. GPT-5-nano, Gemini-3 Flash, and Haiku 4 all clear the quality bar for routine tasks at a fraction of flagship cost.

Cheapest doesn't mean lowest quality. Many production workloads should default to a cheap model and only escalate when a confidence score or output check fails. That two-tier pattern saves real money at scale.

DeepSeek-V3 and Qwen-2.5-72B are the cheapest strong models if you self-serve. Together AI and Fireworks both host them at very competitive per-token rates.

The ranking

  1. #1OpenAI

    GPT-5 nano

    OpenAI's smallest GPT-5 variant, built for ultra-low-cost classification, routing, and high-volume inference.

    Full specsTry GPT-5 nano
    Context
    400K tokens
    Output · 1M
    $0.40 / 1M tokens
    Modalities
    text, vision

    Why it ranks here. Lowest-cost OpenAI model with vision support. Fast P99 latency. Tracked weakness: Visible quality gap on open-ended tasks.

  2. #2Google

    Gemini 3 Flash

    Google's high-speed, low-cost mid-tier with the same massive context window, popular for high-volume RAG.

    Full specsTry Gemini 3 Flash
    Context
    1.0M tokens
    Output · 1M
    Pricing not published
    Modalities
    text, vision, audio, video

    Why it ranks here. 1M-token context at mid-tier price. Very fast, good for interactive UX. Tracked weakness: Reasoning quality below Pro.

  3. #3OpenAI

    GPT-5 mini

    GPT-5's mid-tier sibling, most of the quality at a fraction of the price, ideal for high-volume production workloads.

    Full specsTry GPT-5 mini
    Context
    400K tokens
    Output · 1M
    $2.00 / 1M tokens
    Modalities
    text, vision, audio

    Why it ranks here. Excellent price-quality ratio for production workloads. Fast first-token latency. Tracked weakness: Quality gap vs. flagship visible on hard reasoning.

  4. #4Anthropic

    Claude Haiku 4

    Anthropic's smallest 4-tier model, fast and cheap with the family's signature tone.

    Full specsTry Claude Haiku 4
    Context
    200K tokens
    Output · 1M
    Pricing not published
    Modalities
    text, vision

    Why it ranks here. Fast and cheap for an Anthropic model. Inherits Claude's sensible defaults. Tracked weakness: Quality gap visible on creative tasks.

  5. #5DeepSeekOpen weights

    DeepSeek-V3

    DeepSeek's flagship 671B-parameter MoE, frontier-level quality at a tiny fraction of frontier prices.

    Full specsTry DeepSeek-V3
    Context
    128K tokens
    Output · 1M
    $1.10 / 1M tokens
    Modalities
    text

    Why it ranks here. Frontier-level quality at open-weight prices. MIT license, clean commercial use. Tracked weakness: No native vision support.

  6. #6AlibabaOpen weights

    Qwen2.5-72B

    The previous-generation Qwen flagship, still widely deployed for stability.

    Full specs
    Context
    128K tokens
    Output · 1M
    Pricing not published
    Modalities
    text

    Why it ranks here. Mature deployment. Apache-2.0. Tracked weakness: Superseded by Qwen3 for new builds.

How to choose

Don't pick on the headline ranking alone. Run your top two picks on a representative sample of your own workload and compare. The numbers in this list are sound, but task-specific quality varies in ways no benchmark fully captures. The criteria above are the right axes to evaluate on, but the weighting depends on your stack.

  • Cost-sensitive workloads, start with the cheapest of the top three; escalate only if quality is the bottleneck.
  • Privacy-sensitive workloads, filter to open-weight picks above. They're labeled with a green badge.
  • Latency-sensitive workloads, see the Fastest LLMs list, which can override task-specific picks.

Frequently asked

  • What is the best model for cheapest llms?
    Our #1 pick is GPT-5 nano from OpenAI. OpenAI's smallest GPT-5 variant, built for ultra-low-cost classification, routing, and high-volume inference.
  • How are these rankings determined?
    We rank by the criteria listed at the top of this page: Output price per 1M tokens; Quality floor, must clear basic instruction-following; Latency. Where two models are close, we prefer the one with stronger production deployment evidence at the time of writing. Read the full methodology for our standards.
  • GPT-5 nano or Gemini 3 Flash?
    Both are top-tier picks. GPT-5 nano edges ahead on the criteria most relevant to this task. Gemini 3 Flash is the strongest alternative, see the head-to-head comparison page for full deltas.
  • Are open-source models on this list?
    Yes where they're competitive. Each entry below shows whether the model ships open weights and under what license.
  • How often is this list updated?
    Weekly. New launches that affect the ranking get reflected within seven days. The "last updated" stamp at the top of the page reflects the most recent dataset commit.

Related guides

Friday digest

The week's AI launches, in your inbox.

One short email every Friday, new models, leaks, and quietly-shipped APIs you missed.