Claude vs OpenAI vs Gemini — Cost Calculator (2026)
Compare 8 frontier LLM models on cost for your specific workload. Anthropic Claude (Haiku/Sonnet/Opus 4.5), OpenAI (GPT-4o-mini/4o/o1), Google (Gemini Flash 2.0/Pro 2.0). Input monthly workload once → see all 8 monthly costs side by side. Includes cache + Batch API discounts. Client-side, no signup, dated 2026-04. For Claude-specific workload optimization, use the Claude cost calculator. For verified case studies, see /case-studies.
Full breakdown — sorted by tier + price
| Model | Total / mo | USD | KRW | Context | Cache | Batch |
|---|---|---|---|---|---|---|
| ● Gemini Flash 2.0 Google | $3.00 | in: $1.50 out: $1.50 | ₩4,140 | 1000K | ✓ | ✓ |
| ● GPT-4o-mini OpenAI | $6.00 | in: $3.00 out: $3.00 | ₩8,280 | 128K | ✓ | ✓ |
| ● Claude Haiku 4.5 Anthropic | $36.00 | in: $16.00 out: $20.00 | ₩49,680 | 200K | ✓ | ✓ |
| ● Gemini Pro 2.0 Google | $50.00 | in: $25.00 out: $25.00 | ₩69,000 | 2000K | ✓ | ✓ |
| ● GPT-4o OpenAI | $100 | in: $50.00 out: $50.00 | ₩138,000 | 128K | ✓ | ✓ |
| ● Claude Sonnet 4.5 Anthropic | $135 | in: $60.00 out: $75.00 | ₩186,300 | 1000K | ✓ | ✓ |
| ● o1 OpenAI | $600 | in: $300 out: $300 | ₩828,000 | 128K | — | — |
| ● Claude Opus 4.5 Anthropic | $675 | in: $300 out: $375 | ₩931,500 | 200K | ✓ | ✓ |
How to read this
Models are grouped by tier (small / mid / large) so you compare apples to apples. Within each tier, we sort cheapest first.
- Small tier: Haiku 4.5, GPT-4o-mini, Gemini Flash 2.0 — for high-volume classification, summarization, simple Q&A.
- Mid tier: Sonnet 4.5, GPT-4o, Gemini Pro 2.0 — general production use, code, agents.
- Large tier: Opus 4.5, o1 — complex reasoning, deep code review, strategic decisions.
Within each tier, costs vary. Use case matters more than cost alone — Claude Sonnet 4.5 has 1M context (8x GPT-4o's 128K), Gemini Pro 2.0 has 2M context, but only Claude has Files API + Claude Code + first-party agent SDK.
Quick decision matrix
| If your priority is... | Best choice (2026) |
|---|---|
| Lowest cost on simple high-volume tasks | Gemini Flash 2.0 ($0.075 / $0.30 per MTok) |
| Long context (1M+ tokens) | Gemini Pro 2.0 (2M) or Claude Sonnet 4.5 (1M) |
| Code generation quality | Claude Sonnet 4.5 (SWE-bench leader) |
| Reasoning / math / planning | OpenAI o1 or Claude Opus 4.5 |
| First-party agent + tooling | Claude (Agent SDK + Claude Code, no equivalent at OpenAI/Google) |
| Korean language quality | Claude Sonnet 4.5 (9.4/10 self-measured) or Gemini Pro |
Frequently Asked Questions
Why does cache and Batch matter so much?
Claude API has aggressive prompt caching (90% off cached input) and Batch API (50% off entire job). Real production workloads use these. Without modeling them, raw price comparisons mislead.
What about OpenAI's prompt caching?
OpenAI added prompt caching in late 2024 for GPT-4o (50% off cached input, less aggressive than Anthropic's 90%). This calculator models 50% off OpenAI caches. Anthropic's 90% remains the steeper discount for cached portions.
Why isn't Llama / Mistral / DeepSeek included?
Self-hosted open models have very different cost structures (infrastructure-dependent, not per-token API). They warrant their own comparison page (TBD). This page focuses on managed APIs.
Can I migrate from OpenAI to Claude easily?
Yes — see OpenAI to Claude migration guide. The SDK shapes are similar; main differences: Claude doesn't have response_format: json_object (use tool use instead), and Claude prefers system prompts over messages-array system roles.
Are Bedrock / Vertex prices the same as Direct API?
Roughly. Bedrock typically charges 10-15% premium over Anthropic Direct. Vertex AI for Gemini matches Google AI Studio pricing. Production teams with AWS or GCP commitments often use Bedrock/Vertex for billing consolidation despite the premium. See Bedrock vs Direct.