Claude vs Gemini API pricing

Tier by tier

• Top models: Claude Opus 5.5 $4 / $20 vs Gemini 4 Argon $2 / $10. Sample month (30M input, 6M output): $240 vs $120, so Opus costs 2x as much today. From Dec 1 Argon is $4 / $20, so the sample month becomes $240 on both. Claude Fable 5.1 sits above both at $10 / $50 ($600 a sample month). Head to head: Claude Opus 5.5 vs Gemini 4 Argon.
• Mid tier: Claude Sonnet 5.5 $2 / $10 vs Gemini 4 Argon at its launch price, $2 / $10: the same until Nov 30, then Argon costs twice as much. Gemini 3.1 Pro Preview is $2 / $12 for prompts up to 200K. Head to head: Claude Sonnet 5.5 vs Gemini 4 Argon.
• Small and cheap: Claude Haiku 5.5 $0.10 / $0.50 vs Gemini 3.8 Flash $0.75 / $3.75 and Gemini 3.5 Flash-Lite $0.30 / $2.50. Sample month: $6 vs $45 and $24. On LMArena, Haiku is #14 and Flash #15 for coding, but Flash is well ahead for agents (#11 vs #18), and ranks #5 for chat, where Haiku is not ranked. Haiku's price is for prompts up to 100K tokens; longer prompts cost $0.50 in and $2.50 out. Head to head: Claude Haiku 5.5 vs Gemini 3.8 Flash.

A caching-heavy coding day

Agents and coding tools resend the same context over and over, so cached input matters more than the headline input price. Example: 20M input tokens, 16M of them served from cache, plus 1M output tokens.

At list prices, prompts under each model's long-prompt threshold. Argon from Dec 1 uses Google's $0.20 cached rate. Leaves out Claude cache writes (billed when content first goes into the cache: $5 per 1M on Opus 5.5, $2.50 on Sonnet 5.5) and Gemini cache storage fees, so real bills run a bit higher on both.

Things the headline price hides

• Thinking tokens: Google's output price includes thinking tokens, so a reasoning-heavy Gemini reply costs more than its visible length suggests.
• Long prompts: Haiku 5.5 jumps to $0.50 in, $2.50 out above 100K prompt tokens, and Gemini 3.1 Pro Preview charges $4 in, $18 out above 200K. Opus 5.5, Sonnet 5.5, Argon and 3.8 Flash list one rate.
• Caching: Anthropic bills cache writes as well as cheap cache reads (prices are for the 5-minute cache). Gemini 4 Argon supports implicit caching only; Flash models with explicit caching also pay hourly storage ($0.50 per 1M tokens per hour on 3.8 Flash).
• Batch: both halve the price for batch jobs. Anthropic says "Save 50% with batch processing", and Google's Batch rate for Argon is $1 in, $5 out right now.
• Free tier: Google's free tier covers Gemini 3.8 Flash and 3.5 Flash-Lite free of charge, with usage limits, and Google uses free-tier content to improve its products. Argon is paid only. Anthropic's pricing page lists no free API tier.
• Web search: Claude web search is $10 per 1,000 searches plus tokens. Gemini includes 5,000 Google Search grounding requests a month (shared across Gemini 3 and newer), then $14 per 1,000.
• Price changes ahead: Gemini 4 Argon goes to $4 / $20 from Dec 1; Gemini 3.8 Flash goes to $1.50 / $7.50 from Jan 1, 2027, per Google's pricing page. Claude prices carry no end date.

Which to pick

• High volume, cost first (classification, extraction, simple chat): Claude Haiku 5.5 is the cheapest of the eight at list price, as long as prompts stay under 100K tokens. Gemini 3.8 Flash costs more but ranks higher for agents and chat, and has a free tier for testing.
• Coding and agents: Claude Opus 5.5 leads both LMArena's Code Arena (#1) and Agent Arena (#1) among tracked models; Sonnet 5.5 is #3 for coding at half the price. Argon is #8 for coding.
• Chat, math and roleplay: Gemini 4 Argon ranks #1 for chat, #1 for math and #1 for roleplay, and is the cheaper top model until Nov 30.
• Long documents: Claude Fable 5.1 is #1 on Document Arena, but at $10 / $50 it is the priciest model here.
• Daily personal use: a $20 plan may beat the API. Check Claude Pro vs the API and Google AI Pro vs the API, or compare the apps in Claude vs Gemini plans.

Weighing other providers too? See OpenAI vs Anthropic API pricing, OpenAI vs Gemini API pricing, the cheapest LLM API ranking, or load any two models into the model compare builder.