Model comparisons
Claude Haiku 5.5 bills 80% less than Grok 4.3 per million output tokens. Tier prices, request fields and limits side by side.
Claude Haiku 5.5 chat from Anthropic, released 7 October 2026, through an OpenAI-compatible endpoint. It takes up to 1,000,000 tokens of text and images, returns up to 128,000, and accepts up to 128 tool definitions with forced or automatic tool calls.
View model details →View sample outputs →Grok 4.3 is xAI's chat model behind an OpenAI-compatible call, taking text and image parts in one message array. It reads 1,000,000 tokens, replies up to 900,000, and offers five reasoning-effort levels including none.
View model details →View sample outputs →Compare
Start with the input you have, the result you want and your budget. These are the differences that affect that choice.
Compare the published examples, then try the same input on both models with matching settings.
The Spicy Index and Freedom Score for both models, from the same daily calculation as the leaderboards.
| Measure | Claude Haiku 5.5 | Grok 4.3 |
|---|---|---|
| Spicy Index | 76.5 | 77.0 |
| Freedom Score | 77.8 | 98.9 |
| Evidence | Tested 5 of 5 probe types · 24 runs | Tested 5 of 5 probe types · 24 runs |
| Range if the rest pass | 77.8–77.8 | 98.9–98.9 |
| Price | $0.20 per 1M mixed tokens (3 input : 1 output) | $1.56 per 1M mixed tokens (3 input : 1 output) |
| Reviews | Read the Claude Haiku 5.5 review | Read the Grok 4.3 review |
Current price
A start price answers a different question than the one you have. These are the tiers both models publish, matched setting for setting, in US dollars per billed unit with no conversion step in between.
Swipe horizontally to compare all fields.
| Tier | Claude Haiku 5.5 | Grok 4.3 | Difference |
|---|---|---|---|
| Input | $0.1 / 1 million tokens | $1.25 / 1 million tokens | Claude Haiku 5.5 −92% |
| Cached input | $0.1 / 1 million tokens | $0.2 / 1 million tokens | Claude Haiku 5.5 −50% |
| Output | $0.5 / 1 million tokens | $2.5 / 1 million tokens | Claude Haiku 5.5 −80% |
Arithmetic on the published rates at each model's own default setting, not a quote. The console prices your actual settings before a request is accepted.
About this comparison
Every row here is read from the JSON Schema the endpoint validates against, so it matches what the API will really accept rather than a feature list written by hand.
Swipe horizontally to compare all fields.
| Controls and limits | Claude Haiku 5.5 | Grok 4.3 |
|---|---|---|
| Supported tasks | chat | chat |
| Input context | 1,000,000 tokens | 1,000,000 tokens |
| Maximum output | 128,000 tokens | 900,000 tokens |
| Tool calling | Yes | Yes |
| Execution | Synchronous, with streaming | Synchronous, with streaming |
| Request timeout | 60 minutes | 15 minutes |
| Availability | Live | Live |
Do low-filter models allow every request?
Two separate facts sit behind these labels. The first is what we do, which is nothing: no platform content filter, and no maturity flag to set on a request. The second is what each model does on its own, and that is the only part that varies between these two.
Claude Haiku 5.5
We add no content filter to this model, but like most chat models it declines a blunt request that comes with no context. It writes unrestricted fiction reliably once your system prompt sets up a fiction-writing context, or once the request is framed as a story. Legal and consent duties remain with the customer.
Grok 4.3
We add no content filter to this model, but like most chat models it declines a blunt request that comes with no context. It writes unrestricted fiction reliably once your system prompt sets up a fiction-writing context, or once the request is framed as a story. Legal and consent duties remain with the customer.
Usage note Use only authorized material. Do not invade privacy, impersonate a real person, or create sexual content involving minors.
Use these content labels to shortlist models, then open each model page for available examples and content-test details.
The label never changes your obligations. Age, consent, real-person likeness and the law where you operate stay with you, and SpicyAPI does not inspect each request.
Can I swap the model ID without changing the request body?
The task lifecycle is shared; the body is not. Each endpoint publishes its own JSON Schema and the API validates against that, so this is the list to reconcile before a switch.
How to choose
Grok 4.3 can return more in one reply: up to 900,000 tokens against the other model's ceiling.
Compare
The model identifiers are anthropic/claude-haiku-5.5/chat and xai/grok-4.3/chat, and both go in the same field of the same request.
Both return the response on the chat request. With streaming enabled, read the stream until it finishes; there is no media task to poll.
Both use the same submit-and-track flow, but their required inputs may differ. Check each model page before switching.
One key, one balance, one set of spend caps. Both bill to the same account in US dollars, and the per-key daily, monthly and lifetime limits apply to whichever one you call.
Questions developers ask
Claude Haiku 5.5 costs 80% less per million output tokens. Compare the input and cache rates below for longer conversations. Prices may change; check the estimate shown for your actual settings before submitting.
Claude Haiku 5.5 allows 60 minutes per task and Grok 4.3 allows 15. Those are ceilings rather than typical times — most requests finish far sooner — but they are what your own client timeout has to survive.
Prices and capabilities shown here come from the current public model catalogue. Check each model page for its latest inputs and examples.