ChatGPT alternatives
By Daniel Reeve · tested and updated September 2026
Users searching for OpenAI API alternatives usually do not only want a chatbot. They want model access, pricing control, latency, rate limits, data terms, fallback models, structured outputs and a way to avoid depending on one provider.
Quick answer
Do not pick an API on the input price, which is the number everyone quotes. Output is two to five times more expensive everywhere, and it is where your bill actually comes from.
Rate by rate
| Area | Useful for | Watch out for |
|---|---|---|
| Anthropic Claude API | Strong for long-context reasoning, writing, analysis and careful business outputs. | Check current model availability, rate limits, tool support and enterprise terms. |
| Google Gemini API | Useful for multimodal work, Google ecosystem fit, large context and developer integrations. | Feature names, regions and limits can vary across Google surfaces. |
| Mistral API | Relevant for teams evaluating European AI providers, fast models and model diversity. | Test your own prompts instead of assuming benchmark performance transfers. |
| DeepSeek API | Often considered for coding, reasoning and cost-sensitive workloads. | Privacy, region, reliability and policy fit need careful review before sensitive use. |
| Open-source or local models | Can improve control, privacy and cost at scale. | You own hosting, monitoring, evaluation, security and updates. |
| Routing layers | Let teams switch or compare providers behind one interface. | A routing layer adds another dependency and contract to review. |
Verified API pricing
Per million tokens, read from each provider's own pricing page on 2026-09-10. Output tokens are two to five times the input rate everywhere, which is why a chatty prompt is cheap and a chatty answer is not.
| Model | Input / 1M | Output / 1M | Provider | Source |
|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 | OpenAI | docs |
| GPT-5.6 Sol | $4.00 | $20.00 | OpenAI | docs |
| GPT-5.6 Terra | $2.00 | $12.00 | OpenAI | docs |
| GPT-5.6 Luna | $0.20 | $1.20 | OpenAI | docs |
| Claude Fable 5.1 | $10.00 | $50.00 | Anthropic | docs |
| Claude Opus 5 | $5.00 | $25.00 | Anthropic | docs |
| Claude Sonnet 5 | $2.00 | $10.00 | Anthropic | docs |
| Claude Haiku 4.5 | $1.00 | $5.00 | Anthropic | docs |
| Grok 4.6 | $2.00 | $6.00 | xAI | docs |
Model the bill properly
Estimate your monthly output tokens first, because that is where the money goes. Then apply the two discounts most teams forget: batch processing at roughly half rate for anything not interactive, and prompt caching at around 90% off input for any large fixed preamble.
Only after those two should you compare vendors. A cached, batched workload on a premium model is frequently cheaper than a naive one on a budget model.
Sources
Every figure on this page was read from the official documentation below on 2026-09-10. Prices, limits and model names change without notice — the source is authoritative, this page is not.
FAQ
On published list prices, GPT-5.6 Luna at $0.20 in / $1.20 out per million tokens is the cheapest option in this table, with Claude Haiku 4.5 at $1.00/$5.00 and Grok 4.6 at $2.00/$6.00 next (checked 2026-09-10). Cheapest per token and cheapest per finished task are different questions, though: a weaker model that needs two attempts is not a saving.
Because generating tokens costs the provider more compute than reading them. Practically, it means you should optimise for shorter answers, not shorter prompts. A long, well-specified prompt that produces a tight answer is usually cheaper than a vague one that rambles.
If your work is not interactive — overnight classification, bulk summarisation, backfills — yes, and roughly halving the bill is a large lever for a small change. If a human is waiting for the answer, no.
Roughly 90% off the input rate on cached tokens. For any workload with a large fixed preamble — a system prompt, a document, a schema — reused across many calls, this is normally the single biggest cost reduction available, and it costs nothing but restructuring the call order.
Bursty personal use suits a subscription; steady programmatic volume suits the API. The crossover is not a fixed number, so estimate your monthly tokens, multiply by the output rate above, and compare against the seat price before you commit either way.
No. OpenAI documents a 10% uplift for regional processing endpoints on models released on or after 5 March 2026. If your compliance position requires data residency, price it in before you compare providers.