Three major model releases landed within days of each other in July 2026. Here is how Claude Opus 5, Kimi K3, and Gemini 3.6 Flash actually compare.
By Quantara Editorial Team · AI Tools · Published · 9 min read
Between July 16 and July 24, 2026, three significant model releases landed in quick succession: Moonshot AI's open-weight Kimi K3, Google's Gemini 3.6 Flash, and Anthropic's Claude Opus 5. Each targets a different point on the price-versus-capability curve, and choosing between them depends heavily on the task.
Quick Comparison
| Factor | Claude Opus 5 | Kimi K3 | Gemini 3.6 Flash |
|---|---|---|---|
| Release date | Jul 24, 2026 | Jul 16, 2026 (weights Jul 27) | Jul 21, 2026 |
| Weights | Closed | Open | Closed |
| Input price / 1M tokens | $5 | $3 (cache miss), $0.30 cached | $1.50 |
| Output price / 1M tokens | $25 | $15 | $7.50 |
| Context window | Not fixed by Anthropic on the product page | 1,000,000 tokens | 1,000,000 tokens |
| Standout result | 3x next-best on ARC-AGI 3 | 1st on Frontend Code Arena | ~280 tok/sec, 17% fewer output tokens vs 3.5 Flash |
Claude Opus 5: The Coding and Agent Choice
Opus 5 lands within about 0.5% of Fable 5's peak coding benchmark score at half the price, and reports roughly triple the next-best result on ARC-AGI 3. It is the strongest closed option here for demanding agentic and reasoning-heavy work, especially for teams already inside the Claude ecosystem using Claude Code or Claude Cowork.
Kimi K3: The Open-Weight, Cost-Efficient Choice
Kimi K3's 2.8-trillion-parameter mixture-of-experts design and first-place Frontend Code Arena result make it the strongest open-weight option in this comparison, at output pricing less than a third of Fable 5's. Teams that need to self-host, fine-tune, or keep data fully in-house should weigh Kimi K3 first.
Gemini 3.6 Flash: The High-Volume, Low-Latency Choice
Gemini 3.6 Flash is not trying to compete on frontier reasoning depth. Its combination of the lowest price in this comparison, roughly 280 tokens per second throughput, and 17% fewer output tokens per answer makes it a strong default for high-volume production workloads where cost and latency matter more than peak capability. Note that Google has not yet shipped an updated Gemini 3.5 Pro, so this comparison does not include Google's top reasoning tier.
How to Choose
- Choose Opus 5 for demanding coding, computer-use, and agent tasks where peak capability matters more than raw cost.
- Choose Kimi K3 for open-weight requirements, self-hosting, or the lowest cost per output token among frontier-class options.
- Choose Gemini 3.6 Flash for high-volume, latency-sensitive production traffic where a top-tier reasoning model is not required.
Build Prompts That Work Across All Three
Use Verbito's AI Prompt Generator to write model-neutral prompts with a clear role, task, constraints, and output format, so the same instructions carry over whether you route work to Opus 5, Kimi K3, or Gemini 3.6 Flash.
Sources
- Anthropic: Introducing Claude Opus 5
- Interconnects: Kimi K3 — the open-weights escalation
- TechCrunch: Google releases three new Gemini models, but no 3.5 Pro
Frequently asked questions
- Which model is cheapest: Opus 5, Kimi K3, or Gemini 3.6 Flash?
- Gemini 3.6 Flash has the lowest list price at $1.50 input and $7.50 output per million tokens. Kimi K3 is next at $3 input ($0.30 cached) and $15 output. Claude Opus 5 is the most expensive of the three at $5 input and $25 output per million tokens, though still half of Fable 5's pricing.
- Which of these models is open weight?
- Kimi K3 is the only open-weight model among the three; Moonshot AI released full weights following its July 16, 2026 announcement. Claude Opus 5 and Gemini 3.6 Flash are both closed, hosted-only models.
- Which model should I use for coding?
- Claude Opus 5 and Kimi K3 are both strong coding options: Opus 5 lands near Fable 5's peak coding score at half the price, while Kimi K3 ranks first on the Frontend Code Arena benchmark. Gemini 3.6 Flash is better suited to high-volume, lower-complexity tasks than frontier coding work.
- Is this comparison missing Gemini's best model?
- Yes. Google has not shipped an updated Gemini 3.5 Pro as of this comparison; the company says it is testing one with partners. This article compares Gemini 3.6 Flash, Google's current default model, rather than an unreleased Pro-tier model.