Nexith Core
General-purpose text model: chat, writing, summarization, code, and vision understanding
Model ID: nexith-core · Alias nexith-8b
| Performance |
Speed |
Price |
| High (difficult tasks are escalated automatically) |
Standard |
$8 / $30 per 1M tokens (input/output) |
| Input text, images |
Output text |
Context 1,048,576 tokens |
nexith-core is Nexith's only publicly offered text model. It accepts text and image input, outputs
text, and supports streaming, tool calling, and reasoning-effort control. For difficult tasks, the
gateway escalates internally and invisibly to a stronger upstream, but the public model ID, billing
basis, and identity guardrails all remain nexith-core — callers do not need to (and cannot)
choose the internal upstream.
Pricing#
| Item |
Price |
| Input |
$8.00 / 1M tokens |
| Cached input |
$0.80 / 1M tokens (10% of the input price) |
| Output |
$30.00 / 1M tokens |
- Billed from actual
usage: non-streaming response bodies include usage; for streaming you must send
stream_options: {"include_usage": true} to receive the usage frame before [DONE].
- Image inputs are converted to upstream tokens and counted on the input side; there is no additional per-image charge.
- When the balance is insufficient, the request returns
402 insufficient_balance before reaching the upstream.
Modalities#
| Modality |
Support |
| Text |
Input + output |
| Images |
Input only (image_url, supporting https URLs, data URLs, nx-file://<id>) |
| Audio |
Not supported |
| Video |
Not supported |
Endpoints#
| Endpoint |
Description |
POST /v1/chat/completions |
Primary endpoint, OpenAI-compatible, with Nexith extension fields |
POST /v1/responses |
Responses adapter (Codex, etc.) |
POST /v1/messages |
Anthropic Messages adapter (Claude Code) |
POST /v1/completions |
Legacy completions (kept for compatibility only) |
Capabilities#
| Capability |
Supported |
Notes |
| Streaming |
✅ |
SSE; see Streaming protocol |
| Tool calling |
✅ |
tools / tool_choice / parallel_tool_calls |
| Structured output |
⚠️ |
Achieved via tool calling or prompting; no strict response_format constraint |
| Reasoning effort |
✅ |
reasoning_effort: none/low/medium/high |
| Vision understanding |
✅ |
Images as image_url content blocks |
| Long-term memory |
✅ |
Nexith extensions inject_memory / return_memory / remember |
| Web search |
✅ |
Nexith extensions nexith_web_search / nexith_deep_research |
| Image-generation intent detection |
✅ |
Nexith extension nexith_detect_media |
Model ID and alias#
| ID |
Description |
nexith-core |
Stable ID, recommended |
nexith-8b |
Compatibility alias; behaves identically to nexith-core |
Rate limiting#
Bucketed by (API key, model), resolved in this order:
- Key-level override (the console can set it per key)
- Model-level default (
models.rpm_limit / tpm_limit)
- Plan default (Free/Pro/Enterprise, configured by operations)
- System fallback: 60 RPM / 40,000 TPM
Exceeding the limit returns 429 with X-RateLimit-Limit-Requests,
X-RateLimit-Remaining-Requests, X-RateLimit-Reset, and Retry-After. See
Errors and rate limiting for details.