Summarize a 50-page research paper, and tasks like it, ranked by this site's own recorded cost, speed, and accuracy, blended equally. Not one hand-picked answer: the top 3, and why each one placed where it did.
MiniMax M3 is the best overall blend of cost, speed, and accuracy for this task: 88% accurate, 2.3 s to run, $0.00.
Gemma 4 31B is close behind — the cheapest of the three here: 77% accurate, 3.7 s to run, $0.00.
Nemotron 3.5 Lightning is close behind — the cheapest of the three here: 78% accurate, 3.5 s to run, $0.00.
These get talked about a lot for this category, but we haven’t run them through our own tests yet, so we won’t rank them next to a model we’ve actually measured. Worth knowing exists, not a recommendation.
Every pick above is backed by this site’s own recorded runs against the live Anthropic, OpenAI, Meta, DeepSeek, Google, Z-AI, NVIDIA, and MiniMax APIs; never a marketing number. How we measure → · Full model comparison → · New models →