# Grok 4.7 Costs a Quarter as Much as Claude and GPT-6, and It Shows > Source: > Published: 2026-09-24 13:58:25+00:00 *xAI priced Grok 4.7 at roughly a quarter of what Claude and GPT-6 charge, and on independent coding tests it scores about half as well, a gap that tracks the discount almost exactly.* xAI launched Grok 4.7 on September 21, 2026, calling it the company's most capable model yet for coding, agentic tasks, and general knowledge work. It runs $2 per million input tokens and $6 per million output tokens. Claude Fable 5.1 and GPT-6 Astra both charge $10 and $50 for the same pair. On output tokens, the ones that actually drive most bills, Grok 4.7 costs a little over a tenth as much. That's the pitch. A 500,000-token context window, tool calling, web and X search, code execution, and a price tag built to undercut two of the most expensive frontier models on the market. For a founder deciding which API to wire into a coding agent, the sticker price alone makes Grok 4.7 hard to ignore. Then you look at what it actually does. On the Artificial Analysis Intelligence Index, the independent benchmark suite that averages ten separate evaluations including Terminal-Bench, SciCode, and Humanity's Last Exam, Grok 4.7 scores 46. Claude Fable 5.1 and GPT-6 Astra both land at 53. Artificial Analysis said the score puts xAI back among the top four AI labs, up two points from Grok 4.6. It's also seven points behind the two models it's supposed to be undercutting. [Grok 4.5 doesn't win on benchmarks but it wins on the number founders actually pay](https://startupfortune.com/grok-45-doesnt-win-on-benchmarks-but-it-wins-on-the-number-founders-actually-pay/) xAI's Grok 4.5 places fourth on Artificial Analysis's Intelligence Index but costs roughly $2.49 per coding task versus $10 or more for Claude Opus 4.8. The token efficiency gap is forcing founders to rethink how they choose AI models for their products. - [grok 4.5 api pricing comparison](https://startupfortune.com/grok-45-doesnt-win-on-benchmarks-but-it-wins-on-the-number-founders-actually-pay/) - [cheaper ai model for startups](https://startupfortune.com/grok-45-doesnt-win-on-benchmarks-but-it-wins-on-the-number-founders-actually-pay/) The real damage shows up in the exact category xAI is selling this model for. On Terminal-Bench 4.0, the agentic coding test that measures whether a model can actually complete real command-line tasks, not just talk about them, Grok 4.7 scores 26 percent. GPT-6 Astra hits 60 percent. Claude Fable 5.1 hits 55 percent. Both rivals score more than double Grok's number, on the one benchmark that's supposed to be Grok 4.7's headline strength. Run through xAI's own Grok Build coding harness instead of the standard test setup, Grok 4.7's Terminal-Bench score climbs from 18 percent to 33 percent, according to xAI's own released figures. That's real improvement over Grok 4.6. It still leaves the model roughly half of where Claude and GPT-6 sit on the same task, just measured a different way. There's a second problem hiding under the sticker price, and it's the one that should worry a developer more than the raw accuracy gap. According to Artificial Analysis, Grok 4.7 used about 81,000 output tokens per task on the Intelligence Index run, more than double what Grok 4.6 used and roughly three times the token count GPT-5.6 Sol needed to answer the same questions. A model that's a third the price per token but burns three times the tokens per task isn't actually a third the price. It's close to even, before you even account for the answer being less likely to be right. That's the trap in comparing frontier models on sticker price alone. Cheap tokens are only cheap if the model doesn't need three times as many of them to finish the job, and doesn't need a human to redo the work once it gets there. None of this makes Grok 4.7 a bad model. A 46 on the Intelligence Index and a top-four ranking from Artificial Analysis is a genuine result, and the jump on EEBench, from 53 percent to 64 percent, is real progress in a specific domain. xAI also moved its Harvey Legal Agent Benchmark score from 15.8 percent to 19.6 percent, a category where it still trails badly but is at least climbing. The problem is narrower than "Grok is worse." It's that xAI is marketing this specific release as a frontier coding agent, at a price built to pull developers away from Claude and GPT-6, on exactly the benchmark where it loses worst. Frankly, a quarter of the price for half the reliability isn't a bargain if you're the one debugging what the agent got wrong. For a solo developer running a few scripts, Grok 4.7's price might be worth the tradeoff. For anyone building a production coding agent meant to run unattended, the math looks a lot less generous than the invoice suggests. **Also read:** [Google DeepMind's New Chief Says Gemini 4 Could Ship Much Sooner Than Planned](https://startupfortune.com/google-deepminds-new-chief-says-gemini-4-could-ship-much-sooner-than-planned/) • [OpenAI Fires Contractors Caught Using ChatGPT to Rate ChatGPT's Answers](https://startupfortune.com/openai-fires-contractors-caught-using-chatgpt-to-rate-chatgpts-answers/) • [Microsoft Slashes Copilot Prices for Its Biggest Buyers as a Super App Looms](https://startupfortune.com/microsoft-slashes-copilot-prices-for-its-biggest-buyers-as-a-super-app-looms/) [OpenAI launches GPT-6 Sol and Luna at half the price the same week rivals ship too](https://startupfortune.com/openai-launches-gpt-6-sol-and-luna-at-half-the-price-the-same-week-rivals-ship-too/) OpenAI launched GPT-6 Sol and Luna on September 22 with API prices cut roughly 50% from the GPT-5.6 line, landing the same week Anthropic shipped Opus 5.5 and xAI released Grok 4.7. All three labs are now competing as much on price as on benchmarks, a shift that changes the economics for any founder building on top of these models. - [openai launches GPT-6 Sol and Luna models pricing](https://startupfortune.com/openai-launches-gpt-6-sol-and-luna-at-half-the-price-the-same-week-rivals-ship-too/) - [new GPT-6 models half price competitor releases week](https://startupfortune.com/openai-launches-gpt-6-sol-and-luna-at-half-the-price-the-same-week-rivals-ship-too/) *This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.* ## Join the discussion [Open in the community →](https://startupfortune.com/community/) Almost there. Sign in and your reply posts straight away.