Head-to-head
Gemini 3 Pro vs Kimi K3: which AI model wins in 2026?
Gemini 3 Pro ($12/1M out (prompts ≤200K)) and Kimi K3 ($15/1M out) are two of the most-used AI models in 2026. Across 6 community votes, Gemini 3 Pro leads with 57% approval.
Quick verdict
On Reasoning, Gemini 3 Pro and Kimi K3 are tied at 4.5/5. On budget, Gemini 3 Pro wins: it starts at $12/1M out (prompts ≤200K) versus $15/1M out for Kimi K3.
Line-by-line comparison
Strengths and weaknesses
Gemini 3 Pro
- Topped LMArena at launch with a record 1501 Elo and scored 91.9% on GPQA Diamond, state of the art at release
- ARC-AGI-2 at 31.1%, roughly 6x Gemini 2.5 Pro (4.9%) and nearly double GPT-5.1 (17.6%) at the time
- Best-in-class multimodal understanding: 81% MMMU-Pro, 87.6% Video-MMMU, with a 1M-token context window
- Strong agentic coding: 76.2% SWE-bench Verified, 54.2% Terminal-Bench 2.0, 1487 Elo on WebDev Arena
- Undercut rivals on price at $2/$12 per 1M tokens, below Claude Sonnet-class pricing ($3/$15)
- Configurable thinking_level (low/medium/high) lets developers trade reasoning depth against latency and cost
- Overconfident hallucinations: on AA-Omniscience it gave a wrong answer 88% of the time instead of declining, vs 48% for Claude Sonnet 4.5 (the-decoder)
- Sycophancy widely reported by reviewers (Zvi Mowshowitz: 'vast intelligence with no spine'); needs tight system prompts
- Tool-calling reliability issues in agent stacks: devs reported tool outputs dumped into the chat thread and more scaffolding needed than OpenAI/Anthropic models
- Slow at high thinking level: time to first token measured around 30-60s on AI Studio despite ~130 tok/s output speed
- Retired: shut down on the Gemini API and AI Studio on March 9, 2026, with gemini-3-pro-preview now aliased to Gemini 3.1 Pro
Kimi K3
- #3 on the Artificial Analysis Intelligence Index (57) at launch, comparable to Claude Opus 4.8 and GPT-5.5: the closest a Chinese lab has come to the closed US frontier
- 93.4% on SWE-bench Verified in Vals AI's independent harness (GPT-5.6 Sol: 96.2%, Claude Fable 5: 95.0%) and #1 on Arena.ai's Frontend Code Arena at 1679 points, ahead of Fable 5
- 93.5% on GPQA Diamond (between Fable 5's 92.6% and GPT-5.6's 94.1%), 96.1% on AIME 2025, and Elo 1668 on GDPval-AA v2 agentic work, second only to Fable 5
- Strong agentic profile: #1 on AutomationBench-AA SaaS workflows (53%), long-horizon terminal and repo navigation via Kimi Code, and ~21% fewer output tokens than K2 for more intelligence
- $3/$15 per 1M tokens (input drops to $0.30 on cache hit), flat across the full 1M context: still well under US closed-frontier pricing, with an OpenAI-compatible API and OpenRouter availability
- Native multimodal input (text, image, video) and open weights published on Hugging Face under the MIT-style Kimi K3 License, making it the first 3T-class open-weights frontier model
- 3x price jump over Kimi K2.6 ($0.95/$4 to $3/$15) makes it the most expensive Chinese model ever shipped, at Claude Sonnet 5 list price: the '10x cheaper than US models' era is over (Simon Willison documented the hike)
- Slow: 33 output tokens/s, ranked #145 of 190 models on Artificial Analysis, a poor fit for interactive use
- Hallucination rate climbed from 39% (K2.6) to 51% on AA-Omniscience as the model now attempts answers it would previously refuse (Fable 5 sits at a comparable 54.9%)
- Political censorship on sensitive topics in Mandarin (deflects on Xi, CCP, Tiananmen) and Beijing jurisdiction for API data, a compliance blocker for many EU and enterprise deployments; UK AISI/CAISI also found its cyber guardrails failed to block exploit-development attempts during testing
- 'Open weights' is theoretical for most: ~594 GB in native MXFP4 requiring a multi-GPU datacenter to self-host, and the Kimi K3 License demands a separate deal from API resellers above $20M in yearly revenue
Cast your verdict
One recommendation per tool per gladiator. It reshapes the crowd score everyone sees.
The arena’s verdict on Gemini 3 Pro
A landmark release that put Google back on top in late 2025, with a huge reasoning jump over Gemini 2.5 Pro and the best multimodal scores of its generation. As of mid-2026 there is no reason to choose it: Google shut it down on the API on March 9, 2026, and Gemini 3.1 Pro costs exactly the same while more than doubling ARC-AGI-2 performance (77.1% vs 31.1%). Teams on legacy deployments should migrate to 3.1 Pro, which the old model ID now points to anyway. Avoid it for hallucination-sensitive workloads unless you add grounding, a weakness reviewers flagged repeatedly.
The arena’s verdict on Kimi K3
Kimi K3 is the strongest argument yet that the frontier is no longer exclusively American: #3 on Artificial Analysis, top-tier GPQA and SWE-bench Verified scores, and the best frontend-code arena ranking in the business, at roughly half to a third of US closed-frontier prices. Take it for agentic coding, frontend work and SaaS automation where its benchmarks are strongest, or if your roadmap depends on self-hosting a frontier-class model once the weights land. Skip it for interactive products (33 tokens/s is slow), for anything touching politically sensitive content or strict EU data-residency requirements (Mandarin-language censorship, Beijing jurisdiction), and for high-accuracy retrieval where its 51% hallucination rate on AA-Omniscience demands a verification layer. Cost-obsessed teams should note DeepSeek V4 still delivers vastly more tokens per dollar; K3's pitch is peak capability per dollar, not cheapest tokens.
What the crowd says
On Gemini 3 Pro
“Confidently wrong is its worst mode. On AA-Omniscience it gave a wrong answer 88% of the time instead of declining. Add the sycophancy and you need a tight system prompt to trust it.”
“ARC-AGI-2 at 31% was about 6x Gemini 2.5 Pro and nearly double GPT-5.1 at the time. For visual-heavy work (81 MMMU-Pro) nothing else came close.”
“1501 Elo on LMArena at launch was deserved. Multimodal is where it kills, I feed it lecture videos and dense PDFs and it just gets it. 1M context helps.”
On Kimi K3
“Tried it for our knowledge-base assistant and the hallucination rate is real: it confidently invented two API endpoints in one afternoon. 33 tokens per second on top of that. Back to waiting for the open weights to fine-tune.”
“Our Zapier-style automation stack runs on it now. Tool orchestration that GPT-5.5 fumbled works reliably, and the cache-hit pricing makes repeated workflows genuinely cheap.”
“The Frontend Arena ranking is deserved. Gave it the same dashboard spec I gave Fable 5 and K3's React came out cleaner, with fewer invented props. At $3/$15 through OpenRouter it was a drop-in swap.”
Keep comparing
Frequently asked questions
Is Gemini 3 Pro better than Kimi K3?
The right pick depends on your use case. The line-by-line comparison on this page breaks down pricing, key specs and arena ratings.
Which is cheaper, Gemini 3 Pro or Kimi K3?
Gemini 3 Pro is cheaper: it starts at $12/1M out (prompts ≤200K), while Kimi K3 starts at $15/1M out.
How much do Gemini 3 Pro and Kimi K3 cost per 1M tokens?
Gemini 3 Pro: $2/1M in (prompts ≤200K) per 1M input tokens, $12/1M out (prompts ≤200K) per 1M output tokens. Kimi K3: $3/1M in per 1M input tokens, $15/1M out per 1M output tokens.