Gemini 3.1 Pro Preview vs GPT-5.1

Summary of our 12-test comparison (scores are from our testing):

• Gemini 3.1 Pro Preview wins: structured_output 5 vs 4, creative_problem_solving 5 vs 4, agentic_planning 5 vs 4. These wins indicate Gemini is stronger at JSON/schema compliance, producing non-obvious feasible ideas, and goal decomposition/failure recovery in our tests. Gemini’s structured_output = 5 ranks "tied for 1st with 24 other models out of 54 tested" in our rankings; creative_problem_solving = 5 is "tied for 1st with 7 others out of 54"; agentic_planning = 5 is "tied for 1st with 14 others out of 54."
• GPT‑5.1 wins: classification 4 vs 2. In practice this means GPT‑5.1 is better at routing/categorization tasks and simple label decisions in our tests; classification = 4 places GPT‑5.1 "tied for 1st with 29 other models out of 53 tested."
• Ties (no clear winner in our tests): strategic_analysis (5/5), constrained_rewriting (4/4), tool_calling (4/4), faithfulness (5/5), long_context (5/5), safety_calibration (2/2), persona_consistency (5/5), multilingual (5/5). For example both models scored 5 on long_context (our retrieval accuracy at 30K+ tokens) and both scored 5 on faithfulness (sticking to source material), so neither has a measurable advantage there in our suite. External benchmarks (attributed):
• AIME 2025 (Epoch AI): Gemini 3.1 Pro Preview scored 95.6, ranking 2 of 23 in our records; GPT‑5.1 scored 88.6, ranking 7 of 23. This supports Gemini’s strong math/olympiad performance on that external test (Epoch AI).
• SWE-bench Verified (Epoch AI): GPT‑5.1 scores 68 (rank 7 of 12) in the payload; Gemini has no SWE-bench Verified score provided in our data. Use GPT‑5.1's SWE-bench result as a data point for coding-related real-world GitHub issue resolution (Epoch AI). What this means for real tasks: pick Gemini when you need high-assurance structured outputs, complex planning/agentic flows, or top-tier creative problem generation — its internal 5/5 scores and top ranks in those categories reflect that. Pick GPT‑5.1 if classification and token cost matter more, or if you rely on the provided SWE-bench Verified score for coding benchmarks.

Raw per-MTOK pricing from the payload: Gemini 3.1 Pro Preview charges $2 input / $12 output per 1M tokens; GPT‑5.1 charges $1.25 input / $10 output per 1M tokens. Absolute costs at common volumes (input and output shown separately):

• 1M tokens: Gemini input $2, output $12; GPT‑5.1 input $1.25, output $10.
• 10M tokens: Gemini input $20, output $120; GPT‑5.1 input $12.50, output $100.
• 100M tokens: Gemini input $200, output $1,200; GPT‑5.1 input $125, output $1,000. Example combined scenarios (1:1 input:output ratio):
• 1M in + 1M out/month = Gemini $14 vs GPT‑5.1 $11.25 (Gemini +$2.75/mo).
• 10M in + 10M out/month = Gemini $140 vs GPT‑5.1 $112.50 (Gemini +$27.50/mo).
• 100M in + 100M out/month = Gemini $1,400 vs GPT‑5.1 $1,125 (Gemini +$275/mo). Interpretation: Gemini is ~20% more expensive on output tokens and ~60% more on input tokens (priceRatio in payload = 1.2). Teams with heavy output-volume workloads (content generation, long-form summarization) will see the largest dollar gap and should prefer GPT‑5.1 if cost is the primary constraint. Teams that need Gemini’s strengths (structured JSON outputs, agentic planning, high creative/problem-solving scores) should budget for the higher rates.

Choose Gemini 3.1 Pro Preview if you need: strict JSON/schema compliance, advanced agentic planning, or high creative problem solving (Gemini wins structured_output, agentic_planning, creative_problem_solving in our tests; AIME 2025 95.6, rank 2/23). Choose GPT‑5.1 if you need: lower per-token costs and stronger classification performance (GPT‑5.1 wins classification and is $1.25 input / $10 output per 1M tokens; SWE-bench Verified 68 on Epoch AI). If you operate at high token volume and cost sensitivity is primary, prefer GPT‑5.1; if accuracy of structured outputs or agentic reliability reduces human overhead, the higher token cost for Gemini can be justified.