· CostMyTokens · Analysis · 4 min read
Cheapest LLM API in 2026: Price per Token Ranked
Every LLM API we track, ranked by what 10M tokens a month actually costs — official and gateway prices labeled separately, each with its source and verification date.
Sticker prices per million tokens are only half the story: output tokens cost 4–6x more than input on nearly every model, so the ranking depends on your mix. Below, every model in our dataset is ranked by the cost of one realistic monthly workload, computed from the same verified prices our calculator uses.
Disclosure: links to Novita AI are affiliate links — we may earn a commission at no extra cost to you. It never affects order: every table here is sorted by price only. Read our disclosure.
The ranking: 10M tokens a month
Workload: 10M tokens per month, split 8M input / 2M output (a typical chat or RAG mix), no caching, standard tier. Prices are USD per 1M tokens.
| # | Model | Price type | Input / 1M | Output / 1M | Cost for 10M tokens / month | Source |
|---|---|---|---|---|---|---|
| 1 | OpenAI GPT OSS 120B | Gateway (Novita AI) | $0.05 | $0.25 | $0.90 | Novita AI gateway price (source), verified 2026-10-25 |
| 2 | Claude Haiku 5.5 | Official | $0.1 | $0.5 | $1.80 | official list price (source), verified 2026-10-25 |
| 3 | GPT-6 Luna | Official | $0.1 | $0.5 | $1.80 | official list price (source), verified 2026-10-24 |
| 4 | Llama 4 Maverick Instruct | Gateway (Novita AI) | $0.27 | $0.85 | $3.86 | Novita AI gateway price (source), verified 2026-10-25 |
| 5 | Gemini 3.1 Flash-Lite | Official | $0.25 | $1.5 | $5.00 | official list price (source), verified 2026-10-25 |
| 6 | DeepSeek V4 Pro (off-peak) | Official (DeepSeek, off-peak) | $0.66 | $1.98 | $9.24 | official price (source), verified 2026-10-25 |
| 7 | DeepSeek R1 0528 | Gateway (Novita AI) | $0.7 | $2.5 | $10.60 | Novita AI gateway price (source), verified 2026-10-25 |
| 8 | Gemini 3.8 Flash | Official | $0.75 | $3.75 | $13.50 | official list price (source), verified 2026-10-25 |
| 9 | DeepSeek V4 Pro | Gateway (Novita AI) | $1.32 | $3.96 | $18.48 | Novita AI gateway price (source), verified 2026-10-25 |
| 10 | DeepSeek V4 Pro (peak hours) | Official (DeepSeek, peak hours) | $1.32 | $3.96 | $18.48 | official price (source), verified 2026-10-25 |
| 11 | Qwen3.8 Max | Gateway (Novita AI) | $2.0 | $6.0 | $28.00 | Novita AI gateway price (source), verified 2026-10-25 |
| 12 | Claude Sonnet 5.5 | Official | $2.0 | $10.0 | $36.00 | official list price (source), verified 2026-10-25 |
| 13 | GPT-6.1 Sol | Official | $2.0 | $10.0 | $36.00 | official list price (source), verified 2026-10-24 |
| 14 | Gemini 4 Argon | Official | $2.0 | $10.0 | $36.00 | official list price (source), verified 2026-10-24 |
| 15 | Gemini 3.1 Pro | Official | $2.0 | $12.0 | $40.00 | official list price (source), verified 2026-10-25 |
| 16 | Claude Opus 5.5 | Official | $4.0 | $20.0 | $72.00 | official list price (source), verified 2026-10-25 |
| 17 | GPT-6 Astra | Official | $10.0 | $50.0 | $180.00 | official list price (source), verified 2026-10-24 |
Gateway prices are what Novita AI (affiliate link) charges to host open-weight models; they are not prices set by the model makers. Official prices are what the model’s maker charges for direct API access (OpenAI, Anthropic, Google, and DeepSeek for DeepSeek V4 Pro). DeepSeek bills by time of day: Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday (excluding Chinese public holidays); all other hours are off-peak at half price.
How the answer changes with the mix
The same models at an output-heavy mix (10M tokens, 3M input / 7M output — long generations, code, reports):
| Model | Price type | Cost, 8M in / 2M out | Cost, 3M in / 7M out |
|---|---|---|---|
| OpenAI GPT OSS 120B | Gateway (Novita AI) | $0.90 | $1.90 |
| Claude Haiku 5.5 | Official | $1.80 | $3.80 |
| GPT-6 Luna | Official | $1.80 | $3.80 |
| Llama 4 Maverick Instruct | Gateway (Novita AI) | $3.86 | $6.76 |
| Gemini 3.1 Flash-Lite | Official | $5.00 | $11.25 |
| DeepSeek V4 Pro (off-peak) | Official (DeepSeek, off-peak) | $9.24 | $15.84 |
| DeepSeek R1 0528 | Gateway (Novita AI) | $10.60 | $19.60 |
| Gemini 3.8 Flash | Official | $13.50 | $28.50 |
(Eight cheapest at the output-heavy mix, sorted by that column.)
Cheapest official API
At 10M tokens a month, the cheapest official option is Claude Haiku 5.5 at $1.80 (official list price (source), verified 2026-10-25). Tied at the same price: GPT-6 Luna. Scaling is linear: 100M tokens at the same mix costs $18.00.
Cheapest overall (gateway)
The cheapest option in our dataset is OpenAI GPT OSS 120B on Novita AI at $0.90 for the same workload (Novita AI gateway price (source), verified 2026-10-25). It is an open-weight model, so compare quality for your task before switching — price per token is only meaningful if the model can do the job.
Fine print that changes the ranking
- GPT-6.1 Sol: Prompts above ~272K input tokens enter a long-context pricing tier (about 2x input, 1.5x output). This calculator uses standard-tier rates.
- Claude Haiku 5.5: Prompts over 100,000 tokens are billed entirely at the higher tier ($0.50 input / $2.50 output per 1M) on this model. This calculator uses the base tier.
- Gemini 4 Argon: Introductory pricing runs until Nov 30, 2026. From Dec 1, 2026, rates double to $4 / $20 per 1M.
- Gemini 3.1 Pro: Prices shown apply to prompts up to 200K tokens; longer prompts are billed at a higher tier on this model.
- Gemini 3.8 Flash: Promotional rate through Dec 31, 2026 — this model doubles to $1.50 / $7.50 per 1M from Jan 1, 2027.
- Caching: cached input is billed far below the input rate on many models (e.g. GPT-6.1 Sol at $0.1 per 1M). A workload that reuses a long system prompt can reorder this table — see our prompt caching guide.
- Batch: Anthropic and Google publish batch rates at half price for latency-tolerant work — see batch API discounts explained.
Run your own numbers
These rankings use one fixed mix. Plug your real input, output and cached token counts into the CostMyTokens calculator to rank models for your workload. For DeepSeek’s official peak and off-peak schedule across both models, see the DeepSeek provider page.
Disclosure: links to Novita AI are affiliate links — we may earn a commission at no extra cost to you. It never affects order: every table here is sorted by price only. Read our disclosure.