LLM API Price Comparison 2026: GPT vs Claude vs Gemini
Live prices from the zaimod platform (Nov 2026): GPT-6, Claude Opus, Gemini 3.1, DeepSeek V4 and more, compared side by side with a cost-estimation method.
Price is the most concrete variable in model selection. Using zaimod's live list prices (synced 2026-11-03), this post puts mainstream chat models in one table and gives you a method for estimating the real cost of a feature — so the marketing word "from" stops fooling anyone.
Flagship tier
| Model | Input | Output | Best for |
|---|---|---|---|
| gpt-6-astra | $10.00 | $50.00 | Hard reasoning, long agent chains |
| claude-opus-4-7 | $8.00 | $40.00 | Long documents, code architecture |
| gemini-3.1-pro-preview | $4.00 | $24.00 | Multimodal, long context |
| gpt-5.5 | $7.50 | $60.00 | General flagship |
| claude-opus-5 | $5.00 | $25.00 | Coding, agents |
| gpt-5.6-sol | $6.00 | $30.00 | Reasoning + coding |
| qwen3.7-max | $6.00 | $18.00 | Chinese-first workloads |
Workhorse tier (where most production belongs)
| Model | Input | Output | Notes |
|---|---|---|---|
| gpt-4.1 | $2.00 | $4.00 | Reliable general baseline |
| gpt-5.6-terra | $3.00 | $15.00 | Multimodal + coding |
| claude-sonnet-5 | $4.00 | $20.00 | Best-in-class coding feel |
| gemini-3-pro-preview | $3.00 | $18.00 | Multimodal, long docs |
| deepseek-v4-pro | $12.00 | $24.00 | Strong reasoning, pricey output |
| doubao-seed-2-0-pro-260215 | $5.00 | $24.00 | Chinese business, agents |
| qwen3.5-plus | $0.80 | $5.00 | Excellent Chinese value |
| glm-5.2 | $8.00 | $28.00 | Coding + agents |
Budget tier (batch jobs, support bots, classification)
| Model | Input | Output | Notes |
|---|---|---|---|
| gpt-4o-mini | $0.16 | $0.62 | Most mature small model |
| gemini-3.1-flash-lite-preview | $0.25 | $1.50 | Fast and cheap |
| gemini-2.5-flash | $0.30 | $2.50 | Multimodal workhorse |
| gpt-4.1-nano | $0.30 | $0.60 | Ultralight |
| qwen-flash | $0.32 | $2.00 | Cheap Chinese option |
| doubao-seed-2-0-mini-260428 | $0.50 | $3.00 | Chinese multimodal |
There's also deepseek-v4-flash-free for prototyping (rate-limited — not for production).
Estimating a real feature's cost
Take a support bot: 2,000 tokens average input per turn (system prompt + history), 500 tokens output, 1,000 conversations a day:
- gpt-4o-mini: 2M × $0.16 + 0.5M × $0.62 ≈ $0.63/day
- gemini-3.1-flash-lite-preview: 2M × $0.25 + 0.5M × $1.50 ≈ $1.25/day
- claude-sonnet-5: 2M × $4 + 0.5M × $20 ≈ $18/day
Same feature, 28× cost spread. The right pattern: run classification, summarisation and support on the budget tier; reserve flagships for steps that genuinely need deep reasoning.
Per-call models (image / video)
Image and video models usually bill per call:
- gpt-image-2: $0.06/call
- doubao-seedream-5-0-260128: $0.033/call
- qwen-image-3.0: $0.20/call
- doubao-seedance-2-0-260128: $0.01/call
Three ways to spend less
- Watch the group multiplier. Default group runs at 1.0×; official and premium groups at 1.5×. Default covers every model and is the cheapest unless you specifically need official-channel stability.
- Output costs 3–8× input. Capping max_tokens and trimming system prompts beats model-swapping for immediate savings.
- Cache your system prompt. Keep the static prefix stable so provider prompt caching kicks in — 50%+ savings on repeated context.
Keeping up with price changes
Vendors reprice constantly. Bookmark the zaimod pricing page (it syncs the platform pricing API) or poll https://api.zaimod.com/api/pricing yourself for a cost dashboard.
Start on the budget tier, walk up only when quality demands it. Downgrading later is easy; starting flagship-first and optimising down almost never happens.