Direct answer
Start with deepseek-v4.1-flash-1017 at $0.05 input and $0.15 output per 1M tokens for general text and code. Move bulk workload to qwen3.8-27b, and keep glm-5.3-flash or kimi-k3 only where your prompt set proves they win the task.
Live numbers as of 2026-11-14
These figures come from the public stats snapshot generated on 2026-11-14 from backend ledgers. They show volume and reliability only; they do not prove any single route wins your workload. Run your own 50 prompts before you move budget.
| Metric | Value | What it tells you |
|---|---|---|
| All-time requests | 126,617 | 109 days since launch |
| All-time tokens processed | 5,460,329,009 | 2,299,586,240 served from cache |
| Requests, last 30 days | 116,189 | 5,437,162,745 tokens across 82 active routes |
| Registered developers (all time) | 424 | Self-serve accounts, prepaid |
| Live model routes | 31 | Catalogue at /models; machine list at /v1/models |
| DeepSeek V4.1 Flash 1017, last 30 days | 63,202 | cache-hit rate 12.94% |
| DeepSeek V4.1 Flash Economy (free), last 30 days | 8,741 | cache-hit rate 97.36%; free window, not an SLA |
| 30-day uptime | 98.95% | 88,390 health checks; status currently reports attention, re-check /status before production |
What the routes cost right now
Prices are USD per 1M tokens, taken from the live registry snapshot dated 2026-11-14. The KES columns use the checkout rate of KES 133 = $1; your bank or card issuer may add its own conversion margin on top.
The catalogue is authoritative at /v1/models (the OpenAI-compatible listing) and at /models (the public card with context windows and rate limits). Check both before a production commitment; prices move with provider terms.
See the live list from your terminal: curl -s https://api.fikraapi.co.ke/v1/models -H "Authorization: Bearer YOUR_FIKRA_KEY"
| Model ID | Input / 1M | Cached input / 1M | Output / 1M | Input KES / 1M | Output KES / 1M |
|---|---|---|---|---|---|
| deepseek-v4.1-flash-1017 | $0.05 | $0.0075 | $0.15 | KES 6.65 | KES 19.95 |
| deepseek-v4.1-flash | $0.05 | $0.0075 | $0.15 | KES 6.65 | KES 19.95 |
| qwen3.8-27b | $0.045 | $0.045 | $0.35 | KES 5.98 | KES 46.55 |
| glm-5.3-flash | $0.1875 | $0.0375 | $0.625 | KES 24.94 | KES 83.12 |
| glm-5.3 | $1.225 | $0.225 | $3.85 | KES 162.93 | KES 512.05 |
| kimi-k3 | $2.4375 | $0.24375 | $12.1875 | KES 324.19 | KES 1,620.94 |
Choosing routes in Ghana
| Need | Start with | Watch |
|---|---|---|
| General chat and support | deepseek-v4.1-flash-1017 | 12.94% of input tokens were cache hits last 30 days; tune stable prefixes |
| Structured output and JSON | qwen3.8-27b | temperature 0; validate with a schema check before storing |
| Coding second-eye | glm-5.3-flash | 60.79% cache-hit rate last 30 days; cheap retries while iterating |
| Hard reasoning | deepseek-v4-pro-pinned | $1.65/$4.95 per 1M; confirm availability at /v1/models |
| Very long context | deepseek-v4-flash-pinned or 1017 | cached input $0.00375 and $0.0075 respectively; long prompts reward cache hits |
| Kimi-specific evals | kimi-k3 | $2.4375/$12.1875 per 1M; use for Chinese-context workloads that justify it |
Route math that decides the bill
- Input price is not the bill. Cache-hit rate, output length and retry count decide the wallet cost; the free economy route ran 97.36% cached input in the last 30 days, which is why it looks cheaper than it is.
- Output tokens cost 3x input on flash routes. Cap max_tokens to what the answer actually needs; verbose drafts are the silent budget leak.
- Convert to GHS at your real rate: the Kenya checkout uses KES 133 = $1, and every other market converts at the issuer rate. Compare routes in local currency, not in headline dollars.
- A route that is 10% cheaper per token but fails one task in five is not cheaper. Score accepted tasks, not tokens.
Test protocol before you commit
- Freeze a 50-prompt set from your real production traffic class, including 10 failure-shaped prompts.
- Run it against two or three candidate routes at temperature 0; log request ID, model, tokens and total cost for each.
- Score task success, factuality and wallet cost per accepted task. Keep the pair that wins on accepted-task cost.
- Re-run the set whenever /models lists a route revision; pinned IDs exist precisely for reproducibility.
How to start in Ghana in ten minutes
- Create the account at /account/signup and verify the email address.
- Copy the fk_live_ key from the dashboard and store it somewhere safe; it is shown once.
- Open /dashboard and pay by USD card. The minimum top-up is $5. Conversion from Ghanaian cedi happens at your card issuer's rate and fees, not at a Fikra-set rate. Kenya also shows M-Pesa and Airtel Money; Ghana does not.
- Point your OpenAI client at https://api.fikraapi.co.ke/v1 and set the model to deepseek-v4.1-flash-1017.
- Run 50 prompts from your own workload, then open /stats and the dashboard to check token spend and cache hits before you scale.
- Cap the key with a budget in the dashboard, keep the route that wins your prompt set, and drop the rest.
curl https://api.fikraapi.co.ke/v1/chat/completions \
-H "Authorization: Bearer YOUR_FIKRA_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"deepseek-v4.1-flash-1017","messages":[{"role":"user","content":"Summarise this ticket in two lines."}]}'
Questions buyers ask
Can I pay from Ghana without a foreign credit card?
Not at this checkout. Ghana buyers pay by USD card; the Kenya checkout additionally shows M-Pesa and Airtel Money. Mobile money is not available outside Kenya on this checkout, so plan your funding path before onboarding a team.
How much do I need to start?
The minimum top-up is $5. Self-serve plans are prepaid per 30 days: Lite $10, Pro $25, Max $60. Every tier allows 3,000 requests per minute and 250,000 per day. There is no free production tier; free windows are trials.
Which model ID should I use first?
Use deepseek-v4.1-flash-1017 for text, code and summarisation at $0.05 input and $0.15 output per 1M tokens. Confirm it is listed at /v1/models before production, then compare against qwen3.8-27b and glm-5.3-flash on your own prompts.
Does Fikra sign an SLA?
No. Self-serve is prepaid and has no SLA. Platform 30-day uptime is published at /status (98.95% over 88,390 checks as of 2026-11-14, with status reported as attention). If you need a contractual commitment, contact support before you sign anything.
Where do I see my spend?
The dashboard shows balance and per-key usage; /stats publishes platform-wide telemetry from the same ledgers. Keep request IDs from production calls if you need to reconcile a charge.