FAQGemini, Grok & Other Provider FAQs
Pricing for Google Gemini, xAI Grok, DeepSeek, Meta Llama, Mistral, Perplexity, Azure OpenAI, and Manus.
How much does Google Gemini cost?
Gemini 3.1 Pro Preview, Google's newest Pro model, is $2.00/$12.00 per MTok. The newest Gemini 3.8 Flash costs $0.75/$3.75 on introductory pricing through 31 December 2026, rising to $1.50/$7.50 on 1 January 2027, and Gemini 3.5 Flash-Lite is $0.30/$2.50. The older Gemini 2.5 models cost less (2.5 Pro is $1.25/$10.00 and 2.5 Flash-Lite $0.10/$0.40), but since 18 September 2026 Google has limited them to developers who have used them before and steers new projects to 3.5 Flash-Lite or 3.8 Flash. Google has also deprecated Gemini 2.5 Pro and Gemini 2.5 Flash: from 1 December 2026 their requests go to Gemini 3.8 Flash and Gemini 3.5 Flash-Lite respectively, billed at those models' rates. Gemini models offer 1M-token context windows, 50% batch discounts, and a generous free tier.
How much does xAI Grok cost?
Grok 4.7, the flagship since 21 September 2026, costs $2.00/$6.00 per MTok with a 500K context, rising to $4.00/$12.00 above 200K prompt tokens. That matches the token rates of Grok 4.6 and 4.5 before it; only cached input differs, at $0.50 on 4.7 and 4.6 against $0.30 on 4.5. These are xAI's priciest models, not its cheapest: the earlier flagship Grok 4.3 is $1.25/$2.50 with a full 1M context and an unusually small output premium (2x input, where most providers charge 5x or more). Grok Build 0.1, the coding specialist, is $1.00/$2.00. Batch jobs get a smaller cut at xAI, where most providers give 50%: xAI discounts by only 20%, and then only on Grok 4.7, Grok 4.6, Grok 4.5, Grok 4.3 and the Grok 4.20 variants.
How much does DeepSeek cost?
DeepSeek now prices by time of day. At peak rates, V4.1 Flash costs $0.30/$1.20 per MTok and V4 Pro $1.32/$3.96; off-peak hours bill at half that. Peak runs 01:00–04:00 and 06:00–10:00 UTC on weekdays (9:00–12:00 and 14:00–18:00 Beijing time), and every other hour is off-peak, including all weekend and Chinese public holidays in full. The scheme arrived with a broad price rise on 16 August 2026, and V4.1 Flash then cut Flash rates on 10 September. Cache hits still cost under a cent per MTok. There's no batch API; scheduling work off-peak is DeepSeek's equivalent.
How is Meta Llama priced?
Llama models are open-weight, so you pay a hosting provider (DeepInfra, Together, and others) rather than Meta — and rates vary by host. Representative pricing runs from Llama 3.1 8B at $0.02/$0.04 per MTok to Llama 4 Maverick at $0.27/$0.85 (Novita's rate, after DeepInfra retired Maverick on 1 October 2026). Llama 4 Scout has a 10M-token native context window, the largest we track, though DeepInfra serves 320K of it at $0.10/$0.30.
How does Perplexity Sonar pricing work?
Sonar now runs on Perplexity's Agent API at $0.25/$2.50 per MTok, with web search billed separately at $2.50 per 1,000 calls ($1.00 with Fast Search). That replaces the old bundled per-request search fee: Sonar Chat Completions support ended on 27 September 2026, and synchronous and streaming Sonar calls are now rewritten as Agent API requests. Sonar Pro, Sonar Reasoning Pro and Sonar Deep Research map to Agent API effort levels (called presets until October 2026), which bill the token rates of the underlying OpenAI or Anthropic model plus tool fees. There are no batch discounts.
What does Mistral cost?
Mistral Large 4, Mistral's newest Large model and a public preview since 6 October 2026, is $0.70/$2.10 per MTok. Mistral Large 3, the earlier open-weight 675B flagship, is $0.50/$1.50; Mistral deprecated it on 31 October 2026 and retires it from the API on 30 April 2027. Mistral Small 4 is $0.15/$0.60 and the code-focused Codestral is $0.30/$0.90. Batch jobs get 50% off across Mistral's API.
Is Azure OpenAI cheaper than OpenAI direct?
No — list prices match. Azure GPT-5.5 costs $5.00/$30.00 per MTok and GPT-6 Astra $10.00/$50.00 on Global Standard, the same as OpenAI direct. Teams choose Azure for enterprise agreements, regional data residency, and compliance rather than price. Azure also carries some models at different lifecycle stages, like GPT-5.2 at $1.75/$14.00.
How does Manus credit pricing work?
Manus is credit-based rather than per-token: the free plan gives up to 1,500 credits a month, and paid plans run from $20/month for 4,000 credits to $200/month for 40,000. A typical data-analysis task consumes about 200 credits; a full app build can use 900+.