Agents for humans first
Your agent opens and funds the account. You keep the key.
Fikra API leads with agents for humans (A4H). An agent creates a provisional account for a person, gets a scoped token, funds it by card or mobile money, runs the call, and hands the account over with a claim link. Both payment methods are charged in KES at KES 133 = $1 and credited in USD.
Human inference is the second way in. Sign up directly, get $0.50 in credit, start on the free DeepSeek route, and use the normal OpenAI SDK. Agent-native access is the third: x402 pay-per-call on Base, MCP and A2A on the same account.
We do not claim a proprietary model, a self-serve SLA, SOC 2, or universal zero retention. The trust centre states what is true today.
Machine-readable: pricing.md, agents.json, changelog.json.
TRACK 1
Agents for humans
An agent creates a provisional account for a person, gets a 6-hour scoped token, creates keys, funds it by card or mobile money, and hands over a claim link. Agent-created accounts start with no signup credit by design.
TRACK 2
Human inference
Sign up directly, get $0.50 in credit including the free DeepSeek route, generate a key, and call the OpenAI-compatible API. Start with /authentication.
TRACK 3
Agent-native
x402 pay-per-call, MCP, A2A, API keys, terminal-first flows, no browser and no UI required.
Prices, limits and providers
The current price table lives at fikraapi.co.ke/pricing and is generated from the same registry that answers GET /v1/models. This page no longer keeps a second copy, because the second copy is what drifted.
- Unpaid accounts: 30 requests/minute, 300 requests/day, 100k tokens/minute, 1M tokens/day.
- Paid accounts: 3,000 requests/minute, 250k requests/day, no token ceilings.
- New accounts are prepaid. Existing subscriptions are honored.
- Cached input can never cost more than fresh input.
Core Architecture & Compatibility
Our infrastructure is engineered to strictly adhere to the OpenAI v1 REST specification. You only need to update your base URL and authenticate with your Fikra API Key to achieve a 100% drop-in replacement for existing pipelines.
| Specification Parameter | Production Value | Implementation Detail |
|---|---|---|
| Base URL | https://api.fikraapi.co.ke/v1 | The global endpoint for all inference routing. |
| Authentication Scheme | Authorization: Bearer | Requires active `fk_live_...` key generation. |
| SDK Compatibility | OpenAI SDK v1.0+ | Native support for Python, Node.js, Go, and REST. |
The Engine Matrix
The Fikra engine matrix lists every chat model published on GET /v1/models.
This table is generated live from the model registry: it always matches the API.
Rates are USD per 1M tokens; every model includes automatic high-availability routing behind Fikra.
| Model | Best for | Input / 1M | Output / 1M | Context |
|---|---|---|---|---|
| deepseek-v4.1-flash | $ 0.05 | $ 0.15 | 1,048,576 tok | |
| deepseek-v4.1-flash-1017 | The latest DeepSeek V4.1 Flash update (Oct 17 build): better reasoning, coding, agentic and | $ 0.05 | $ 0.15 | 1,048,576 tok |
| deepseek-v4-flash-0731 | $ 0.05625 | $ 0.17 | 1,048,576 tok | |
| glm-5.3-flash | $ 0.155 | $ 0.49 | 327,680 tok | |
| kimi-k3 | $ 2.1 | $ 10.5 | 1,113,088 tok | |
| step-5-preview | $ 1.25 | $ 3.3 | 1,048,576 tok | |
| mimo-v2.6-pro | Xiaomi MiMo-V2.6-Pro: the top-scoring open-weight model on the Artificial Analysis index rig | $ 0.5 | $ 1.0 | 1,048,576 tok |
| deepseek-v4-flash-ark | DeepSeek V4 Flash served through BytePlus ModelArk. 1M context, 393k output. | $ 0.55 | $ 1.65 | 1,048,576 tok |
| gemini-4-argon | $ 2.5 | $ 12.0 | 1,048,576 tok | |
| glm-5.3 | Z.ai flagship GLM-5.3: large-scale reasoning for complex software engineering and long-horiz | $ 1.05 | $ 3.35 | 327,680 tok |
| glm-5.2 | Z.ai GLM-5.2 on a 1M context: strong coding and reasoning at a lower price than GLM-5.3. The | $ 0.75 | $ 1.65 | 262,144 tok |
| nvidia-nemotron-3.5-super-vl | $ 0.5 | $ 1.25 | 1,048,576 tok | |
| qwen3.8-27b | Fast 27B-class multimodal model with automatic prompt caching. Cached input never costs more | $ 0.045 | $ 0.35 | 1,001,536 tok |
| deepseek-v4-pro-ark | DeepSeek V4 Pro served through BytePlus ModelArk. 1M context, 393k output. | $ 1.65 | $ 4.95 | 1,048,576 tok |
| deepseek-v4.1-flash-ark | DeepSeek V4.1 Flash served through BytePlus ModelArk. 1M context, 393k output. | $ 0.1875 | $ 0.75 | 1,048,576 tok |
| deepseek-v4-1-flash-economy-free | Temporary free promotional route backed by BytePlus ModelArk. Available for two weeks or unt | $ 0.0 | $ 0.0 | 1,048,576 tok |
| fikra-pro-20b | $ 0.2 | $ 0.2 | 131,072 tok | |
| fikra-pro-120b | $ 0.3 | $ 0.3 | 131,072 tok | |
| fikra-qwen-27b | $ 0.045 | $ 0.35 | 262,144 tok | |
| fikra-qwen-max | $ 6.0 | $ 12.0 | 1,048,576 tok | |
| fikra-auto | $ 0.045 | $ 0.35 | 262,144 tok | |
| fikra-auto-pro | $ 6.0 | $ 12.0 | 1,048,576 tok | |
| fikra-code-35b | $ 0.045 | $ 0.33 | 262,144 tok |
Speech and text utilities ship as separate endpoints:
fikra-embed
(embeddings, $0.05/1M tokens),
fikra-stt
(transcription, $0.001/min) and
fikra-tts
(speech, $25/1M characters).
Decoupled Production Scaling
Fikra API employs a Redis-backed Token Bucket algorithm to decouple consumer limits from upstream hardware thresholds. Your throughput capacity (RPM) scales securely based on your account tier, shielding you from backend rate limits.
| Account Tier | Request Limit | Access Condition |
|---|---|---|
| Unverified | 30 RPM · 300 req/day | Default free-evaluation state. Tokens: 100,000 TPM · 1,000,000 TPD. |
| Trusted | 3000 RPM · 250,000 req/day | Unlocked permanently on your first wallet top-up. Tokens: 0 TPM · 0 TPD. |
How do I send my first inference request?
Because we map exactly to the OpenAI spec, you do not need to learn a new library. Simply initialize the client with the Fikra base URL and your `fk_live_` key.