Grok 4.3 launches: xAI's new flagship, 37.5% input / 58.3% output price cuts

Documentation Index

Fetch the complete documentation index at: https://docs.apiyi.com/llms.txt

Use this file to discover all available pages before exploring further.

​

Key Highlights

• xAI’s new flagship — Grok 4.3 (April 30, 2026), positioned for “lowest hallucination rate, strongest agentic tool calling, best instruction following”
• Major price cuts — input down 37.5% and output down 58.3% vs. Grok 4.20 0309 v2; full Intelligence Index eval costs about 20% less
• Blazing fast — 159 tokens/s output, well above the 64.6 t/s median for same-tier reasoning models
• Million-token context + multimodal — 1M context window, text + image input, always-on reasoning (cannot be disabled)
• All common groups available — Default, SVIP, etc. all open; deposit $100 get 10% bonus, ~85% of xAI’s upstream pricing

​

Background

​

Detailed Breakdown

​

Core Capabilities

​

Benchmarks

​

Tech Specs

​

Practical Use

​

Recommended Scenarios

1. Production agent workflows — high-frequency tool calls, multi-step planning, scenarios that demand stable instruction following
2. Large-scale document processing — 1M context + 159 t/s makes long-doc summarization and cross-file audits run fast
3. Customer support / telecom dialogues — 98% on τ²-Bench Telecom suggests strong performance on ticketing and tech-support tool-calling tasks
4. Multimodal analysis — image + text mixed input for screenshot analysis, chart reading
5. Cost-sensitive high-QPS apps — low unit price + fast speed makes Grok 4.3 a reasonable substitute for GPT-5.4 / Claude Sonnet in some flows

​

Code Examples

​

Standard call

from openai import OpenAI

client = OpenAI(
    api_key="your-apiyi-key",
    base_url="https://api.apiyi.com/v1"
)

response = client.chat.completions.create(
    model="grok-4.3",
    messages=[
        {"role": "user", "content": "Analyze this customer ticket and propose a resolution"}
    ],
    max_tokens=4096
)

print(response.choices[0].message.content)

​

Multimodal call (image + text)

response = client.chat.completions.create(
    model="grok-4.3",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What's the cause of this error in the screenshot?"},
                {"type": "image_url", "image_url": {"url": "https://example.com/error.png"}}
            ]
        }
    ]
)

print(response.choices[0].message.content)

​

Agent tool calling

tools = [
    {
        "type": "function",
        "function": {
            "name": "query_order_status",
            "description": "Look up a customer order status",
            "parameters": {
                "type": "object",
                "properties": {"order_id": {"type": "string"}},
                "required": ["order_id"]
            }
        }
    }
]

response = client.chat.completions.create(
    model="grok-4.3",
    messages=[{"role": "user", "content": "Check the status of order #A1024"}],
    tools=tools,
    tool_choice="auto"
)

​

Best Practices

1. Pick it for agents — τ²-Bench / IFBench data points to a clear edge in tool calling and instruction following over generic models
2. Mind the 200K context tier — past 200K, input/output prices double; control prompt length on long-doc workflows
3. Always-on reasoning is default — reasoning tokens count against your output bill; even simple tasks incur some reasoning spend
4. Speed-first chat — 159 t/s throughput makes Grok 4.3 great for “stream-and-render” real-time chat UX
5. Stack the recharge bonus — APIYI’s deposit $100 / +10% offer effectively brings cost to ~85% of upstream

​

Pricing & Availability

​

Pricing (Tiered)

​

Price comparison with recent models

​

Stack with site recharge promotions

​

Available models & groups

​

Access

• Site: apiyi.com
• API endpoint: https://api.apiyi.com/v1
• OpenAI SDK compatible — just swap base_url, api_key, and set model to grok-4.3

​

Summary & Recommendations

• You’re already running Grok 4.20 / Claude Sonnet 4.6 / GPT-5.4 mini for agent tasks — evaluate migration first
• High-frequency tool calling, customer-support automation, telecom-style ticketing
• Cost-sensitive, high-QPS real-time chat products
• Workloads needing image input multimodal analysis

• You’re already on GPT-5.5 / Claude Opus 4.7 for the hardest reasoning (Grok 4.3’s intelligence ceiling is lower)
• You depend heavily on OpenAI/Anthropic-specific API behaviors (e.g., OpenAI’s strict function schema)
• You’re stable on GPT-5.4 / Gemini 3 Pro and the migration upside is marginal