OpenAI's GPT Bel Is Unstoppable But Grok 4.7 Is Something Else
OpenAI has deployed an internal model codenamed Bell that has solved over 100 long-standing mathematical problems through recursive self-improvement. SpaceX released Grok 4.7 at a $6 token price, yet early tests reveal the model produces lazy outputs and fails to match the visual precision of GPT-6 Astra Pro. Benchmarks show Grok 4.7 scores 46.3% on software engineering tasks while costing significantly less than GPT 5.6 SOL Max. The new model lags behind competitors in electrical engineering and legal work despite its lower price point compared to Kimi K3.
I don't know how to feel about Grok 4.7...
Grok 4.7 delivers performance comparable to Opus 5.0 while costing approximately half as much per completed task. XAI positions this model as a top-tier option for coding and knowledge work, achieving a 71% score on DeepSuite benchmarks. The model utilizes a 500k token context window and offers significantly lower costs than Fable 5.1 despite similar output efficiency. Analysts note that Grok 4.7 trails Opus 5.0 in raw capability but excels in cost-effectiveness for enterprise workflows.