Model rules
Context windows, output caps, modalities and what each model tends to refuse. Every row carries the date someone last checked it. Cost says only paid, free or local - exact prices change too often for a community page to keep them honest, so check the vendor's pricing page before you budget anything.
12 rows older than four months
| Model | Context | Max output | Cost | Modalities | Good at | Tends to refuse | Checked |
|---|---|---|---|---|---|---|---|
| GPT-5 OpenAI | 400K | 128K | Paid | text, image, audio | Tool calling, structured output, agentic loops | Weapons, malware, explicit sexual content, medical and legal advice presented as… | May 2026 |
| GPT-5 mini OpenAI | 400K | 128K | Paid | text, image | Cheap structured extraction at volume | Same categories, tighter on borderline security questions | May 2026 |
| Claude Opus Anthropic | 200K | 64K | Paid | text, image | Long code review, editing, following long instruction lists | Weapons and malware detail, sexual content involving minors, impersonating a named living… | May 2026 |
| Claude Sonnet Anthropic | 200K | 64K | Paid | text, image | The default for daily work: fast, cheap enough, follows constraints | Same categories as Opus, a bit more cautious around security tooling | May 2026 |
| Claude Haiku Anthropic | 200K | 32K | Paid | text, image | Classification, extraction, high volume | Same policy categories. It turns down ambiguous requests sooner than the bigger two. | May 2026 |
| Gemini Pro Google | 1000K | 65K | Paid | text, image, audio, video | Very long documents, video and PDF sets | Weapons, self-harm instruction, election claims in some regions | May 2026 |
| Gemini Flash Google | 1000K | 65K | Paid | text, image, audio, video | Long input at low cost | Same categories as Pro | May 2026 |
| Grok 4 xAI | 256K | 64K | Paid | text, image | Recent events, informal drafting | Refuses less than the others, on purpose. That helps and hurts. | May 2026 |
| Llama 3.3 70B Meta | 128K | 8K | Local | text | Private data, offline work, fine-tuning | The base weights refuse very little. What you get depends on the instruct tuning you run. | May 2026 |
| Llama 3.2 8B Meta | 128K | 8K | Local | text | Running on a laptop, classification, drafting | Depends on the tuning | May 2026 |
| Mistral Large Mistral AI | 128K | 32K | Paid | text | Fast extraction and classification | Standard safety categories, fairly relaxed on security research | May 2026 |
| Mistral Small Mistral AI | 128K | 32K | Paid | text | High volume classification | Same categories | May 2026 |
The cost column says only whether a model is paid, free to use, or runs on your own hardware. Exact prices change often enough that any number written here would be wrong within months, and a stale price is worse than none - check the vendor's pricing page before you budget anything.
Something out of date here?
A wrong context window sends people down the wrong path. Corrections take a minute and go live after review.