FIKRA API / CHANGELOG

What's new.

Model releases, product announcements, pricing changes, benchmark results, and developer resources. Everything new across the Fikra API platform.

PRODUCT UPDATES

The latest from Fikra

22 updates
Fikra API
Fikra API @fikraapi · 2026-11-04 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Cached input is cheaper now. Public stats are live.

Cached-input prices for every DeepSeek route now sit at the 20% margin floor, and /stats publishes measured reliability, latency and cache hit rates.

Read full update

Cached input is now priced much closer to the real upstream rate. For every DeepSeek route, the cached-input price is set at the 20% margin floor, and the public pricing table reads from that one config source.

What changed

ModelUpstream cachedRetail cachedMargin
deepseek-v4-flash-0731$0.0090$0.0112520%
deepseek-v4-flash-ark$0.0140$0.0175020%
deepseek-v4-pro-ark$0.0440$0.0550020%
deepseek-v4.1-flash$0.0060$0.0075020%
deepseek-v4.1-flash-1017$0.0060$0.0075020%
deepseek-v4.1-flash-ark$0.0030$0.0037520%

Repeated prompt prefixes are billed at the cached rate. Cached input can never cost more than fresh input. If you run coding agents or multi-turn tools, this is the line that matters most.

Public stats

/stats is now live. It publishes aggregate volume, reliability from our own probe and live traffic, latency for cold and cached requests, cache hit rate per model, and a model integrity table. Empty states are explicit. No customer identifiers and no per-user numbers appear there.

Free DeepSeek API

The free DeepSeek API page now shows live availability from the same server-side gate that routes the model, the exact model ID, the free limits, expiry conditions and copy-paste setup for Cline, Roo Code, Open WebUI and the OpenAI SDKs.

Setup guides

The new guides hub has verified setup pages for coding agents and OpenAI-compatible clients.

Get your key See live prices

Fikra API
Fikra API @fikraapi · 2026-10-31 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

MCP now covers 48 fixed-price tools plus account controls

Fikra's MCP server now exposes document, contract, recruitment, developer, creator and meeting tools with free price discovery, alongside account, balance, usage and receipt tools. Same account, same billing.

Read full update

The MCP server now covers 48 fixed-price tools and the account surface. Discovery is free through tools/list and /v1/products. Each paid call returns a price, a receipt id and a usage row. The Agent for Humans flow and existing prices are unchanged.

Write-up: /newsroom/fikra-api-expands-mcp-capabilities

Fikra API
Fikra API @fikraapi · 2026-10-31 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Agents for humans: every paid endpoint, three ways to pay

An agent can open a Fikra API account for a human, mint a key, fund it by checkout link, M-Pesa STK push or saved card, and use chat, images, video, audio, batches and jobs with one token.

Read full update

Agents can now open and fund a Fikra API account for a human, then use every paid endpoint with one delegated token: chat, Seedream images, MiniMax-H3 video, speech, embeddings, batches and fixed-price jobs.

Top-ups default to $10 with a $1 minimum. Pay by checkout link, an M-Pesa prompt that runs in the terminal, or a saved card after the human claims the account. Above $100 needs an explicit confirm.

Full write-up: /newsroom/agents-for-humans

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

mimo-v2.6-pro: 80.4/100 on the Fikra Eval 500 benchmark

mimo-v2.6-pro scored 80.4/100 on Fikra Eval 500.

Read full update

Fikra ran mimo-v2.6-pro through the Fikra Eval 500 (500 tasks). Score: 80.4/100. Average task latency: 41616 ms.

Full results: https://fikraapi.co.ke/benchmark/mimo-v2.6-pro

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

glm-5.3: 76.6/100 on the Fikra Eval 500 benchmark

glm-5.3 scored 76.6/100 on Fikra Eval 500.

Read full update

Fikra ran glm-5.3 through the Fikra Eval 500 (500 tasks). Score: 76.6/100. Average task latency: 73569 ms.

Full results: https://fikraapi.co.ke/benchmark/glm-5.3

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

nvidia-nemotron-3.5-super-vl: 78.0/100 on the Fikra Eval 500 benchmark

nvidia-nemotron-3.5-super-vl scored 78.0/100 on Fikra Eval 500.

Read full update

Fikra ran nvidia-nemotron-3.5-super-vl through the Fikra Eval 500 (500 tasks). Score: 78.0/100. Average task latency: 27648 ms.

Full results: https://fikraapi.co.ke/benchmark/nvidia-nemotron-3.5-super-vl

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

kimi-k3: 79.4/100 on the Fikra Eval 500 benchmark

kimi-k3 scored 79.4/100 on Fikra Eval 500.

Read full update

Fikra ran kimi-k3 through the Fikra Eval 500 (500 tasks). Score: 79.4/100. Average task latency: 49397 ms.

Full results: https://fikraapi.co.ke/benchmark/kimi-k3

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

deepseek-v4-flash-0731: 78.2/100 on the Fikra Eval 500 benchmark

deepseek-v4-flash-0731 scored 78.2/100 on Fikra Eval 500.

Read full update

Fikra ran deepseek-v4-flash-0731 through the Fikra Eval 500 (500 tasks). Score: 78.2/100. Average task latency: 81571 ms.

Full results: https://fikraapi.co.ke/benchmark/deepseek-v4-flash-0731

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

glm-5.2: 78.0/100 on the Fikra Eval 500 benchmark

glm-5.2 scored 78.0/100 on Fikra Eval 500.

Read full update

Fikra ran glm-5.2 through the Fikra Eval 500 (500 tasks). Score: 78.0/100. Average task latency: 94833 ms.

Full results: https://fikraapi.co.ke/benchmark/glm-5.2

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

deepseek-v4.1-flash-1017: 77.8/100 on the Fikra Eval 500 benchmark

deepseek-v4.1-flash-1017 scored 77.8/100 on Fikra Eval 500.

Read full update

Fikra ran deepseek-v4.1-flash-1017 through the Fikra Eval 500 (500 tasks). Score: 77.8/100. Average task latency: 43010 ms.

Full results: https://fikraapi.co.ke/benchmark/deepseek-v4.1-flash-1017

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

qwen3.8-27b: 75.8/100 on the Fikra Eval 500 benchmark

qwen3.8-27b scored 75.8/100 on Fikra Eval 500.

Read full update

Fikra ran qwen3.8-27b through the Fikra Eval 500 (500 tasks). Score: 75.8/100. Average task latency: 47000 ms.

Full results: https://fikraapi.co.ke/benchmark/qwen3.8-27b

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

glm-5.3-flash: 80.2/100 on the Fikra Eval 500 benchmark

glm-5.3-flash scored 80.2/100 on Fikra Eval 500.

Read full update

Fikra ran glm-5.3-flash through the Fikra Eval 500 (500 tasks). Score: 80.2/100. Average task latency: 18890 ms.

Full results: https://fikraapi.co.ke/benchmark/glm-5.3-flash

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

deepseek-v4.1-flash: 74.8/100 on the Fikra Eval 500 benchmark

deepseek-v4.1-flash scored 74.8/100 on Fikra Eval 500.

Read full update

Fikra ran deepseek-v4.1-flash through the Fikra Eval 500 (500 tasks). Score: 74.8/100. Average task latency: 26657 ms.

Full results: https://fikraapi.co.ke/benchmark/deepseek-v4.1-flash

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

step-5-preview: 8.2/100 on the Fikra Eval 500 benchmark

step-5-preview scored 8.2/100 on Fikra Eval 500.

Read full update

Fikra ran step-5-preview through the Fikra Eval 500 (500 tasks). Score: 8.2/100. Average task latency: 546187 ms.

Full results: https://fikraapi.co.ke/benchmark/step-5-preview

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Fikra Eval 500: 11 open models ranked, MiMo leads at 80.4

Fikra ran 500 deterministic tasks across 11 open-model SKUs. MiMo V2.6 Pro leads at 80.4/100.

Read full update

Fikra ran the Fikra Eval 500 across 11 open-model chat SKUs.

Model Score Avg task latency
mimo-v2.6-pro 80.4 41616 ms
glm-5.3-flash 80.2 18890 ms
kimi-k3 79.4 49397 ms
deepseek-v4-flash-0731 78.2 81571 ms
nvidia-nemotron-3.5-super-vl 78.0 27648 ms
glm-5.2 78.0 94833 ms
deepseek-v4.1-flash-1017 77.8 43010 ms
glm-5.3 76.6 73569 ms
qwen3.8-27b 75.8 47000 ms
deepseek-v4.1-flash 74.8 26657 ms
step-5-preview 8.2 546187 ms

MiMo V2.6 Pro leads at 80.4, with GLM 5.3 Flash at 80.2 and Kimi K3 at 79.4.

The benchmark is deterministic and runs 500 tasks across arithmetic, numeric reasoning, knowledge, coding-light, data manipulation, language, extraction and instruction following. It is not a replacement for MMLU, GPQA or SWE-bench.

Full results and per-category scores: https://fikraapi.co.ke/blog/fikra-eval-500-open-models

Live benchmark table: https://fikraapi.co.ke/benchmark

Fikra API
Fikra API @fikraapi · 2026-10-27 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Signup protection: proof-of-work, live email checks and source blocks

Automated signups were creating fake accounts and causing bounces. Signup now requires a proof-of-work challenge and a live email check, and high-confidence fake accounts were removed.

Read full update

What changed

Signup now requires a SHA-256 proof-of-work challenge solved in the browser. The form calls a live email check for syntax, MX/A records, disposable domains and machine-generated local parts. A honeypot and three failed challenges block the source IP and device for seven days across the site and API.

What was removed

High-confidence fake accounts were backed up locally and removed, including their keys, usage rows and profiles. No account with a successful payment was removed.

Why

The bounces were hurting the sending domain. The old checks proved a domain could receive mail, not that a mailbox belonged to a real person.

Details: https://fikraapi.co.ke/blog/signup-abuse-protection

Fikra API
Fikra API @fikraapi · 2026-10-25 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Model settings update: larger output budgets and corrected context windows

We corrected the context windows and maximum output budgets for the DeepSeek, GLM, Kimi, Qwen and MiMo lanes. The gateway now reads these limits from the model registry on every request, so large max_tokens requests are no longer clamped by a stale default.

Read full update

Model settings update: larger output budgets and corrected context windows

We corrected the context windows and maximum output budgets for the main chat lanes. The gateway now reads these limits from the model registry on every request instead of falling back to a stale default. Requests within the published ceiling pass through unchanged.

Model Context Max output
DeepSeek V4.1 Flash 1017 1,048,576 393,216
DeepSeek V4.1 Flash 1,048,576 384,000
DeepSeek V4 Flash 0731 1,048,576 131,072
GLM 5.2 262,144 131,072
GLM 5.3 327,680 131,072
GLM 5.3 Flash 327,680 131,072
Kimi K3 1,113,088 1,048,576
Qwen 3.8 27B 1,001,536 32,768
MiMo V2.6 Pro 1,048,576 131,072

Output plus input cannot exceed the context window. The max output number is the ceiling for one response.

  • Live catalog: https://api.fikraapi.co.ke/v1/models
  • Model page: https://fikraapi.co.ke/models
  • Agent onboarding: https://fikraapi.co.ke/for-agents
Fikra API
Fikra API @fikraapi · 2026-10-23 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Fikra API load test results, 22 Oct

Gateway concurrency and latency numbers for the DeepSeek lineup after the auth and thread-pool fixes.

Read full update

Full load and latency writeup for 22 Oct. Streaming /v1/chat/completions, max_tokens 400, 20 runs per cell, cached rows verified by prompt_tokens_details.

Latency, cold vs warm cache:

model TTFT p50 cold TPS p50 cold TTFT p50 cached TPS p50 cached
deepseek-v4.1-flash 1.77s 86 1.81s 107
deepseek-v4.1-flash-1017 2.02s 102 1.88s 81
deepseek-v4-flash-0731 1.36s (cold outlier) 5 1.92s 82

The 0731 snapshot still wobbles upstream on a cold KV cache. Cached runs settle it to 1.92s and 82 TPS. Full tables on the DeepSeek page.

Concurrency through the gateway, streaming:

  • 10 parallel: 10/10 success.
  • 25 parallel: 25/25 success.
  • 50 parallel: upstream drops most sessions past roughly 40 concurrent. The gateway now holds a provider slot limit of 44 and answers overflow with HTTP 429 and a Retry-After, instead of opening more sockets into a wall.

What changed to get there: auth verification and the balance check moved onto worker threads, a 15 second key cache sits in front of the auth check and purges on revoke, the thread pool grew to 64, and upstream 5xx errors before the first byte walk to the next key automatically.

Billing itemizes cached input per request via prompt_tokens_details, and cached tokens bill at the lower cached rate.

Model catalog and pricing: fikraapi.co.ke/models.

Fikra API
Fikra API @fikraapi · 2026-10-18 Announcement

Fikra API / Update

Product announcement

What's changing across Fikra

↗

Deepseek V4.1 Flash 0731 is now available in Fikra API

Deepseek V4.1 FLah 0731 is a new model from Deepseek with pareto performance improvements

Read full update

Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our API.

🛠️ Optimized for DeepSeekHarness , Tencent CodeBuddy, WorkBuddy, and now Fikra API

Fikra API
Fikra API @fikraapi · 2026-10-16 New model
Model release Fikra catalog

New model available

Available through the Fikra API

↗

NVIDIA Nemotron 3.5 Super VL is now available in Fikra API

A new open model from NVIDIA with 120B total parameters with 12B active

Read full update

Meet NVIDIA Nemotron 3.5 Super VL

A new open model combines reasoning and tool use with text, image, document, and video understanding for long-running AI agents.

120B total parameters with 12B active

Fikra API
Fikra API @fikraapi · 2026-10-16 New model
Model release Fikra catalog

New model available

Available through the Fikra API

↗

Step 5 Preview is now avaiable in Fikra API

Step 5 Preview is a new flagship model from Stepfun AI

Read full update

Introducing Step 5 Preview: Advancing the Pareto Frontier.

Step 5 Preview is a new flagship model from Stepfun AI for agentic work, delivering frontier-level performance across software engineering and professional knowledge work, with particular strength in finance.

  • 600B total / 27B active MoE, with 1M context + Vision
  • Substantially lower task cost at comparable intelligence
  • Broad software engineering capabilities with sustained execution over long horizons
  • Now open weights
Fikra API
Fikra API @fikraapi · 2026-10-13 New model
Model release Fikra catalog

New model available

Available through the Fikra API

↗

Gemini 4 Argon is now available in Fikra API

Gemini 4 Argon, Google's latest frontier model, is now live on Fikra API from day 0

Read full update

Gemini 4 Argon, Google's latest frontier model, is now live on Fikra API from day 0. Use it through the same OpenAI-compatible endpoint you already use, with no new SDK and no separate Google account.

Google built Argon for deep work in complex, real-world tasks: software engineering, finance and legal analysis, cybersecurity defense, and scientific reasoning. It is a new model, not a fine-tune of an earlier one.

Highlights

  • Frontier agentic coding. 77.9% on DeepSWE v1.1 and 91.9% on Vibe Code Bench.
  • Enterprise knowledge work. 68.9% on the Vals Index and 65.4% on Vals Finance Agent v2.
  • Long context that holds up. 84.2% on GraphWalks at 256k to 1M tokens.
  • Deep visual understanding. 91.7% on LVBench and 71.6% on Chartography.
  • Multimodal inputs. Text, images, audio and video in, text out.

Benchmark figures are reported by Google in the Gemini 4 Argon model card. Methodology is published by Google DeepMind.

Model details

Model ID gemini-4-argon
Architecture Sparse mixture-of-experts transformer
Input Text, images [, audio, video]
Output Text
Context window 1M tokens (input + output combined)
Knowledge cutoff March 2026
Input price $2.50 / 1M tokens
Output price $12.00 / 1M tokens
Streaming Yes

Billing works like every other Fikra model: prompt and completion tokens are metered separately and deducted from your USD wallet. M-Pesa and card top-ups both work.

Try it

curl https://api.fikraapi.co.ke/v1/chat/completions \
  -H "Authorization: Bearer $FIKRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-4-argon",
    "messages": [{"role": "user", "content": "Hello, Argon."}],
    "stream": true
  }'

Or with the OpenAI Python SDK:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.fikraapi.co.ke/v1",
    api_key="YOUR_FIKRA_API_KEY",
)

resp = client.chat.completions.create(
    model="gemini-4-argon",
    messages=[{"role": "user", "content": "Hello, Argon."}],
)
print(resp.choices[0].message.content)

Get started

  1. Create an account or log in to your dashboard.
  2. Top up your wallet and generate an API key.
  3. Set model to gemini-4-argon.

You can also try it in the browser at fikraapi.co.ke/models/gemini-4-argon/play before writing any code.

Good to know

  • Like all foundation models, Argon can hallucinate, and you may see occasional slowness or timeouts at launch.
  • Its knowledge cutoff is March 2026, and some domains may be less current.
  • Use is subject to Google's Generative AI Prohibited Use Policy.
  • Questions or issues? Reach us at [email protected].
BUILD WITH FIKRA

Found something worth trying?

Explore available models, experiment in the playground, and build AI applications through one OpenAI-compatible API.