xAI introduced Grok 4.3—a cost-effective AI model focused on agent tasks and enhanced practicality
xAI launches Grok 4.3 – a model that saves on tokens and runs faster
Elon Musk’s xAI company has unveiled a new version of its AI model — Grok 4.3. Compared with its predecessor (Grok 4.20), it reduces the cost of input tokens by 40 % and output tokens by 60 %. This makes it more affordable for developers and businesses, and also increases operational efficiency.
| Metric | Grok 4.3 | Grok 4.20 |
|---|---|---|
| Cost per input token (million $) | 1.25 | 2.08 |
| Cost per output token (million $) | 2.50 | 6.25 |
| Speed | 100 tokens/second | – |
| Context window | 1,000,000 tokens | – |
Key features
* Autonomous search – the model independently searches the internet and X (formerly Twitter).
* Code execution – support for Python scripts.
* RAG (Retrieval-Augmented Generation) – searching local files.
* Document generation – Excel, PDF, PowerPoint.
Eric Jiang, head of development at xAI, noted that when creating Grok 4.3, the main focus was on speed, low cost, and expanded tool capabilities. The model runs at 100 tokens per second and supports a context of up to one million words.
Testing cost
A full run of the model across all Artificial Analysis tests cost only $395, which is significantly cheaper than GPT‑5.5 ($3,959) and Claude Opus 4.7 ($4,811). Experts believe Grok 4.3 has reached a “Pareto frontier” – an optimal price-to-performance ratio.
Intelligence Index scores
* 53 points (higher than the predecessor but lower than OpenAI and Anthropic).
* Compared with Muse Spark and Claude Sonnet 4.6, the model is slightly higher.
* Four points better than Grok 4.20, yet noticeably behind market leaders.
GDPval‑AA test results
* Elo rating increased by 321 points to 1,500.
* Surpassed Google’s Gemini 3.1.
* GPT‑5.5 remains the leader with a margin of 276 points.
Other independent tests
| Test | Result |
|---|---|
| Andon Labs (automation management) | The model sometimes “falls asleep” and fails to complete tasks. |
| Vals AI – CaseLaw, CorpFin | Grok 4.3 took first place. |
| General code and complex math challenges | 13th place. |
New features
xAI also introduced a beta version of Agent Mode for Grok Imagine, available via the web interface. The mode is intended not for single queries but for long-term creative projects: planning, generation, editing, and refinement of materials. According to xAI, the tool is suitable for creating short films or product stories. Enabling Agent Mode is done in the lower left corner of the web interface and requires a paid subscription.
Where to get Grok 4.3
* Through OpenRouter
* xAI API
* Hermes agent
Thus, Grok 4.3 represents an economical and fast model with broad capabilities for developers and businesses, but still lags behind competitors in some specialized tasks.
Comments (0)
Share your thoughts — please be polite and stay on topic.
Log in to comment