Grok 4.5: a new competitor to GPT and Claude
Posting this a little late, but I prepared something interesting. Grok 4.5 is temporarily available to test for free in Grok Build and Cursor.
The model runs at around 80 tokens per second, supports a context window of up to 500,000 tokens, and costs $2 per million input tokens and $6 per million output tokens. According to Musk, Grok 4.5 was trained from scratch on SpaceXAI’s own foundation and is not built on top of Kimi like Composer.
In terms of quality, Grok 4.5 comes close to GPT-5.6 Terra, while its pricing is closer to Luna. Among Anthropic models, it sits between Haiku and Sonnet, making it especially attractive in terms of price-to-performance.
One of the model’s strongest areas is frontend development. Grok 4.5 took third place in Code Arena: Frontend, up from 62nd place for Grok 4.3. It ranked alongside GLM-5.2 Max and above Claude Opus 4.8 Thinking. GPT-5.6 has not yet been added to this leaderboard.
Top comments (0)