xAI Launches Grok 4.6 Model with Performance Approaching GPT-5.6 at Lower Pricing

xAI officially launched the Grok 4.6 model on August 12, 2026, with an input price of $2 per million tokens, $0.5 for cached input, and $6 for output. Prices double once prompt tokens exceed 200,000. The model supports a 500,000-token context window, accepts text and image inputs, and outputs text only with no length limit.

Benchmark Performance and Real-World Gap

According to Artificial Analysis Intelligence Index data, the Grok 4.6 high version is tied with GPT-5.6 Soul, ranking behind Claude Opus 5. On the GDPval benchmark, Grok 4.6 high achieved the highest score, surpassing GPT-5.6 Soul and Claude Opus 5 Max. This benchmark covers real economic tasks including engineering, finance, AutoCAD, hotel management, and video editing.

On the Harvey Bench legal task test, Grok 4.6 scored 15.8%, compared with GPT-5.6 Soul at 2.5% and Claude Opus 5 at 11.3%. On the Cursor Bench 3.2 internal coding test, Grok 4.6 trailed Claude Opus 5 Max. Terminal agent task scores improved from 15% to 26%.

The per-million-token task cost is approximately $0.83, several times lower than that of models with equivalent intelligence.

Release Cadence and Data Sources

Grok 4.5 launched on July 8, 2026, with Grok 4.6 following within four weeks. xAI has confirmed that Grok 4.7 is currently in training, with a target release in three to four weeks. The company acquired coding data through its purchase of Cursor and leveraged idle GPU capacity at the Colossus data center, forming a coding-first loop.

Official release notes indicate that Grok 4.6 is built on the same approximately 1.5-trillion-parameter foundation as Grok 4.5, fine-tuned with engineering data rather than trained as a new foundation model.

Practical Deployment of Grok Bot

The Grok Bot, launched concurrently on August 11, 2026, runs on persistent cloud virtual machines and supports messaging, approvals, connectors, and routine functions. It is positioned for non-technical knowledge workers rather than developers. Users can invoke it directly on the Cursor platform, with positive feedback on coding tasks.

The tool complements the voice and video models previously released in the Grok series, though the specific technical relationship with the SpaceXAI model of the same name has yet to be disclosed.

Industry Impact of the Pricing Strategy

With lower costs for equivalent tasks than comparable OpenAI and Anthropic products, Grok 4.6 directly eases the budget pressure developers face when switching between multiple models. The API supports four reasoning intensity levels—low, medium, high, and xhigh—with high as the default, adjustable to user needs.

This pricing maintains benchmark competitiveness while disrupting the previous pattern of uniformly high prices among frontier models. Developers can verify cost-effectiveness through real terminal tasks rather than relying solely on public leaderboards.

Independent Assessment

Current evidence shows that Grok 4.6 has established a local advantage on specific legal and economic task benchmarks, with pricing and execution speed serving as its primary differentiators.