Agents Directory
SkillsRankingsAgents
CategoriesModelsBenchmarksCompareAgent LeaderboardSkillsRankingsAgentsAbout
/Rankings
/Best cheap AI models

Best cheap AI models

Cheap

You no longer need a $15-per-million flagship for most work. These are the best cheap AI models in 2026: everything here costs at most about $1 per million input tokens, ranked by how much real capability you keep while the price drops.

Which model should you run?

As of June 2026, picking a cheap model is a three-way question:

  • Want the most capability per dollar? DeepSeekDeepSeek V4 or MinimaxMiniMax-M3
  • Already on a big-lab API? GeminiGemini 3 Flash on Google, OpenAIGPT-5.4 Mini on OpenAI, ClaudeClaude Haiku 4.5 on Anthropic
  • Running high volume where every cent counts? DeepSeekDeepSeek V4 Flash, Qwen3.5 Flash, or GLM 4.7 Flash logoGLM 4.7 Flash

One habit worth keeping: cheap models punch above their weight on routine work but slip on long multi-step agent runs. Route your everyday traffic to a cheap pick and keep a flagship for the hard 10%.

The best cheap AI model right now is DeepSeekDeepSeek V4: near-frontier reasoning, a 1M token context window, and open weights at $0.435 per million input tokens, roughly a sixth of what OpenAIGPT-5.4 or ClaudeClaude Sonnet 4.6 cost.

MinimaxMiniMax-M3 is even cheaper at $0.30 per million and scores 54.7 on the Artificial Analysis Intelligence Index logoArtificial Analysis Intelligence Index, within reach of flagships that cost ten times more. From the big labs, GeminiGemini 3 Flash at $0.50 and OpenAIGPT-5.4 Mini at $0.75 are the practical budget tiers.

If you only care about price per token, DeepSeekDeepSeek V4 Flash ($0.098), Qwen3.5 Flash ($0.065), and GLM 4.7 Flash logoGLM 4.7 Flash ($0.06) form the floor where models stay genuinely usable.

The ranking
Updated June 2026
Best cheap models overall

The sweet spot: models that keep most of the capability of a flagship at a tenth or less of the price.

#ModelContextInput
  • 1
    1DeepSeek
    DeepSeek V4DeepSeekThe best capability per dollar: near-frontier reasoning, 1M context, $0.435 per million.
    Context1.049M
    Input$0.435/M
  • 2
    2Minimax
    MiniMax-M3MiniMaxNear-frontier intelligence at $0.30 per million input tokens.
    Context205K
    Input$0.3/M
  • 3
    3Gemini
    Gemini 3 FlashGoogle DeepMindGoogle's fast workhorse: 1M context at $0.50 per million input tokens.
    Context1.049M
    Input$0.5/M
Cheapest from the big labs

OpenAI's and Anthropic's budget tiers: pricier than the Chinese picks, but the easiest drop-in if you are already on those APIs.

#ModelContextInput
  • 1
    1OpenAI
    GPT-5.4 MiniOpenAIThe cheapest GPT that still handles real agent work, at $0.75 per million.
    Context400K
    Input$0.75/M
  • 2
    2Claude
    Claude Haiku 4.5AnthropicAnthropic's budget tier: dependable quality at $1 per million input tokens.
    Context200K
    Input$1/M
High-volume and pipeline picks

For when price per token is the whole point: agents that run all day, batch jobs, and sub-agent steps.

#ModelContextInput
  • 1
    1MoonshotAI
    Kimi K2.5Moonshot AIOpen-weight agentic coder at $0.40 per million input tokens.
    Context262K
    Input$0.375/M
  • 2
    2DeepSeek
    DeepSeek V4 FlashDeepSeekA 1M context window at $0.098 per million. The high-volume default.
    Context1.049M
    Input$0.09/M
  • 3
    3Minimax
    MiniMax M2.5MiniMaxOpen weights and solid quality at $0.15 per million input tokens.
    Context205K
    Input$0.15/M
  • 4
    4Qwen
    Qwen3.5-FlashQwen1M context at $0.065 per million, for summarization and routine steps.
    Context1M
    Input$0.065/M
  • 5
    5GLM 4.7 Flash logo
    GLM 4.7 FlashZ.AIThe absolute floor that still works: $0.06 per million input tokens.
    Context203K
    Input$0.06/M
Intelligence vs. price

Each model's Artificial Analysis Intelligence Index logoArtificial Analysis Intelligence Index score against its blended price per 1M tokens. Toward the top right is more intelligence per dollar.

Full interactive leaderboard on our Intelligence Index page.

Frequently asked questions
What is the best cheap AI model in 2026?

DeepSeekDeepSeek V4 is the best cheap AI model right now: near-frontier reasoning, a 1M token context window, and open weights at $0.435 per million input tokens. MinimaxMiniMax-M3 at $0.30 per million is the runner-up with an Intelligence Index score of 54.7, remarkable at that price.

What is the cheapest AI model that is still good?

GLM 4.7 Flash logoGLM 4.7 Flash at $0.06 per million input tokens and Qwen3.5 Flash at $0.065 are the cheapest models that stay genuinely usable, and DeepSeekDeepSeek V4 Flash at $0.098 adds a 1M token context window. All three suit high-volume pipelines, summarization, and sub-agent steps rather than hard reasoning.

How much cheaper are these than GPT-5.5 or Claude Opus?

Flagships run $2.50 to $15 per million input tokens; everything on this list costs at most about $1, and the high-volume picks sit under $0.10. In practice that is a 5x to 100x price gap. Output tokens widen it further, since flagship output prices reach $15 per million while budget models charge $0.20 to $5.

Are cheap models good enough for AI agents?

For routine agent steps, yes: file edits, summaries, structured extraction, and short tool chains work well on models like DeepSeekDeepSeek V4 and OpenAIGPT-5.4 Mini. Long multi-step runs with heavy tool use are where cheap models slip. A common setup routes everyday steps to a cheap model and escalates hard planning to a flagship.

Share:
Details:

  • Models

    10

  • Filter

    Cheap

  • Updated

    June 2026
Browse:SkillsRankingsModelsBenchmarksProvidersAgentsAgent LeaderboardCompareCategories
Quick Links:AboutBlog

© 2026 Agents Directory