Also Read: Programming and Development
TL;DR
Released: September 21, 2026
Model ID: grok-4.7
Price: $2 input / $6 output per million tokens (unchanged). A faster variant costs double.
Size: reportedly 2.1 trillion parameters, with extra training on SpaceX engineering data
Where: Grok app, xAI API, Cursor, Grok Build, GitHub Copilot, and third-party routers — no waitlist
Verdict: a solid upgrade that still sits a step behind Claude Fable 5.1 and GPT-6 Astra
What changed from Grok 4.6
xAI built Grok 4.7 on a larger base model and ran a longer reinforcement-learning phase focused on tasks that take hours rather than a single reply. The company says the model:
spends more time on hard problems,
checks its own answers more often, and
manages long contexts better.
It also ships with what xAI calls an entirely new safeguard stack.
Grok 4.7 benchmarks
Here are the headline numbers xAI and early coverage reported:
Benchmark | Grok 4.6 | Grok 4.7 | Notes |
|---|---|---|---|
CursorBench 4.0 | 40.4% | 46.3% | Longer-running coding tasks |
DeepSWE v1.1 (high effort)
| 65.2% | 71.0% |
|
EEBench (electrical engineering) | 53.0% | 64.0%
| Second only to GPT-6 Astra |
Terminal-Bench 4.0 | — | 38.0% | Fable 5.1 Max scored 57.9% |
AA-Briefcase (Elo)
| ~1546 | 1657 | Second to Fable 5.1 |
On Artificial Analysis's GDPval, Grok 4.7 also placed second behind Claude Fable 5.1. On safety, xAI says it tops LatchBio's biosafety benchmark and blocked all but 3.3% of risky prompts on its internal HackerBench v0.3.
How Grok 4.7 compares
Decrypt summed it up well: Grok 4.7 is a meaningful improvement, but it lands in the middle of the pack. On Cursor's cost-versus-accuracy chart, it costs more per task than GPT-6 Astra and Claude Sonnet 5, while Fable 5.1 wins at every price point.
Also Read: Meta Muse Zero-Day: How a Researcher Hijacked Meta's AI Agent, and What It Means for You
Musk himself had set expectations before launch, saying Grok 4.7 would be roughly on par with Claude Opus 5.0. He's also teased the roadmap: Grok 4.8 as a meaningful step up, Grok 4.9 aimed at the Astra/Fable class, and Grok 5 as a possible frontier leader. None have dates.
Worth noting: Anthropic released Claude Opus 5.5 the very next day at 20 per million tokens, claiming Fable-level performance, which puts fresh pressure on Grok's positioning.
Why developers should care
xAI's strategy is clear: good enough, cheap, and everywhere. At 6, Grok 4.7 is a strong option for:
high-volume coding assistance where top-tier accuracy isn't critical,
apps already integrated with X or Tesla ecosystems,
failover setups where you want a cheaper secondary provider.
xAI's API is OpenAI-compatible, so you can often point an existing OpenAI client at https://api.x.ai/v1 and change the model name:
$client = OpenAI::factory()
->withApiKey(env('XAI_API_KEY'))
->withBaseUri('api.x.ai/v1')
->make();
$response = $client->chat()->create([
'model' => 'grok-4.7',
'messages' => [['role' => 'user', 'content' => 'Explain Laravel queues in 3 sentences.']],
]);
echo $response->choices[0]->message->content;FAQ
When was Grok 4.7 released? September 21, 2026.
How much does Grok 4.7 cost? $2 per million input tokens and $6 per million output tokens — the same as Grok 4.6.
Is Grok 4.7 better than Claude? It trails Claude Fable 5.1 on most published benchmarks but is significantly cheaper.
Where can I use Grok 4.7? The Grok app, xAI API, Cursor, Grok Build, GitHub Copilot and third-party model routers.
Sources: Decrypt, xAI, SQ Magazine, Android Headlines, Basenor, Superpower Daily, CellCog.
