All promptsTech

Find the best AI model for your budget

Here's how to find the best AI model for your budget with ChatGPT or Claude: the top 10 models by intelligence, with price, coding scores and the best value pick.

  1. 01

    Add Firecrawl to your AI

    Some sources ask you to accept their terms first, in Alexandria in your dashboard.

  2. 02

    Copy the prompt

    @Firecrawl Use Alexandria to rank the 10 smartest AI models on Artificial Analysis by price per million tokens, coding scores and time per task, and tell me which is the best value for coding.

    Open in ChatGPT with this prompt
  3. 03

    What came back

    Blended price per 1M tokens

    Artificial Analysis list prices at a 3:1 input-to-output ratio, from the answer below

    $0$10$20$20$2$4Gemini CarbonGPT-6 AstraSonnet 5.5Gemini ArgonKimi K3.1

    Claude Code (Claude Opus 5.5) sent back, word for word:

    GPT-6.1 Terra (Max) is the best value for coding. It costs $2 per 1M tokens, a quarter of Claude Opus 5.5's price. It scores 55.6% on Terminal-Bench 4.0 and 42.3% on FrontierCode 2.0, and it finishes tasks fastest. I pulled this from Artificial Analysis through Firecrawl Alexandria today (Nov 4, 2026), using Intelligence Index v5.0.1 and the highest-scoring effort level for each model.

    #ModelIntelligence IndexBlended $/1M tokensTerminal-Bench 4.0FrontierCode 2.0Time per task
    1Gemini 4 Carbon (High)60.0$865.2%40.6%1,768 s
    2Claude Opus 5.5 (Max, Default Fallback)59.1$859.6%46.4%1,549 s
    3GPT-6 Astra (Max)56.9$2059.1%42.4%1,093 s
    4GPT-6.1 Sol (Max)54.6$456.1%40.6%1,205 s
    5Claude Sonnet 5.5 (Max, Default Fallback)54.5$463.6%41.7%1,835 s
    6GPT-6.1 Terra (Max)51.3$255.6%42.3%761 s
    7Gemini 4 Argon (High)50.4$857.1%25.7%2,041 s
    8Muse Wonder Full (Max)50.1not listedn/an/an/a
    9Claude Fable 5.1 (Xhigh, Default Fallback)50.0$2055.1%35.7%1,158 s
    10Kimi K3.1 Pro46.5$467.2%25.5%2,251 s

    What it means:

    • Terra is the cheapest coder per task. One Terminal-Bench 4.0 task costs $1.49 on GPT-6.1 Terra (Max), compared with $1.82 on GPT-6.1 Sol (Max), $10.05 on Gemini 4 Carbon (High) and $13.11 on Claude Opus 5.5. Its FrontierCode score is within 4.1 points of the leader.
    • If you want top coding scores, pay up. Claude Opus 5.5 (Max) leads FrontierCode 2.0 at 46.4%, and Kimi K3.1 Pro leads Terminal-Bench 4.0 at 67.2%. Kimi costs $4 per 1M tokens, but it is the slowest at 2,251 s per task and scores 25.5% on FrontierCode.
    • Lower effort settings cost less. GPT-6.1 Sol (Xhigh) scores 52.5 on the index at $0.76 per index task, and GPT-6.1 Terra (High) scores 42.4 at $0.26.

    Limits: The blended price assumes 3 input tokens for every 1 output token, at each model's list price. Artificial Analysis returned no output-speed (tokens per second) figures for any model, so "time per task" here is the average time it took to run one Intelligence Index task, not raw generation speed. Muse Wonder Full has no published price or detailed scores. I can also compare API providers for one model, such as cheaper or faster hosts for GPT-6.1 Terra.

    Sources usedRan Nov 4, 2026

More prompts

See all 34 prompts