AI model speed & cost comparison

Averaged across all 25 tasks on this site, for all 14 models. Same method as every other page here: recorded runs where they exist, clearly-labelled placeholder profiles where they don't; never a marketing number.

Is Claude faster than ChatGPT?

Averaged across every task and every model on each side: OpenAI's models come out faster overall (386× vs 521× human speed, Anthropic vs OpenAI). Accuracy averages close, 88% vs 88%, and cost averages $0.05 vs $0.0099 per task. But that average hides real variation: the fastest single model here is GPT-5.6 Luna (808× on average), and tier-for-tier the gap narrows or flips.

Averages are a starting point, not a verdict: Compare AI Models lets you race any specific Claude model against any specific GPT model on one task and see the real numbers for that exact pairing. For a per-category breakdown instead of one blended average, see Claude vs ChatGPT: which is faster? →

Fastest AI model overall

Ranked by average speed multiplier over a typical person, across all 25 tasks:

  1. GPT-5.6 Luna (OpenAI): 808× faster than a person on average, 82% average accuracy
  2. Haiku 4.5 (Anthropic): 740× faster than a person on average, 78% average accuracy
  3. Gemma 4 31B (Google): 678× faster than a person on average, 77% average accuracy

Cheapest AI model overall

Ranked by average cost per task (same tasks, same token-count method, different price per token per model):

  1. Gemma 4 31B (Google): $0.00 average per task, $0.00 total across all 25 tasks
  2. GLM-5.2 (Z-AI): $0.00 average per task, $0.00 total across all 25 tasks
  3. Nemotron 3 Ultra (NVIDIA): $0.00 average per task, $0.00 total across all 25 tasks

The free-tier models here genuinely cost $0.00 per task, and the most expensive model averages roughly 515× the cost of the cheapest paid one (DeepSeek V3.2), for the same set of tasks: that gap is the entire reason this site shows tokens and cost on every result, not just speed. Budgeting for schoolwork specifically? See cheapest AI for students →. Notice an open-weight model near the top of that list? See paid vs open-source AI → for the full breakdown.

Most accurate AI model overall

Ranked by average accuracy against each task's scoring rubric:

  1. Fable 5 (Anthropic): 95% average accuracy, 135× average speed
  2. Opus 4.8 (Anthropic): 92% average accuracy, 240× average speed
  3. GPT-5.6 Sol (OpenAI): 92% average accuracy, 259× average speed

The fastest model and the most accurate model are rarely the same one; see the FAQ for why speed, cost, and accuracy are three separate facts about a model, not one score.

Best model per difficulty tier

This site's own recommendation logic, per tier and provider (the same "good fit" judgment shown on every result):

Tier
Anthropic pick
OpenAI pick
Simple
Haiku 4.5
GPT-5.6 Luna
Intermediate
Sonnet 5
GPT-5.6 Terra
Complex
Opus 4.8
GPT-5.6 Sol

All figures are averages of this site's own per-task numbers: recorded monthly where a run exists, a clearly-labelled placeholder profile otherwise. See how we measure →. Want one specific pairing instead of an average? Discover AI Speed races one task at a time; Compare AI Models runs two models head-to-head. Coding specifically? Fastest AI for coding → Curious what's newest? New models →