xAI (SpaceXAI)
Grok 4.7
xAI's newest flagship, priced and speed-matched to Grok 4.6. Independent coverage is thin: Humanity's Last Exam (43.1% at xhigh effort, versus Grok 4.6's 42.9% at high effort, not a same-tier comparison), LiveBench (77.4, level with Grok 4.6's 78.0), vals.ai's Terminal-Bench 2.1 (73.41, xhigh, a tie with Grok 4.6's 78.28) and AA-AnalystAgent (52.5 pass^5, xhigh). Its DeepSWE figure is a vendor claim.
Grok 4.7 benchmarks and pricing, every number sourced: 6 tracked Grok 4.7 benchmark scores (5 independently run, 1 still resting on a vendor’s own claim), priced at $2.00 per million input tokens and $6.00 per million output.
- Released
- 2026-09-21
- License
- proprietary
- Context window
- 500K tokens
- Knowledge cutoff
- 2026-05
- Verified
- sources checked 2026-09-23–2026-10-15
- Succeeded
- Grok 4.6 (2026-08-12)
- Parameters
- Not disclosed
- Architecture
- Not disclosed
Grok 4.7’s verified record
Grok 4.7’s most-compared rival is GLM-5.3-Flash: 2 leads and 3 not callable across their 5 shared comparisons. Grok 4.7 is priced at $2.00/$6.00 per 1M tokens (in/out) vs GLM-5.3-Flash’s $0.15/$0.50.
Against the 171 head-to-head comparisons Grok 4.7 shares with other tracked models:
A gap counts for Grok 4.7 only where both sides were run independently and the benchmark still separates models. Losses are listed alongside wins on purpose — Grok 4.7 trails on 9 of them.
HLE · no tools Reasoning
±2 is noiseLiveBench Composite score across 7 domains
±2.7 is noiseAnalystAgent Spreadsheet & document analysis
±11.2 is noiseTerminal-Bench 2.1 Terminal ops
±10.6 is noiseTerminal-Bench 4.0 · xhigh Terminal ops
±12.4 is noiseNo verdict for Grok 4.7 anywhere on DeepSWE ( nothing independently confirmed on both sides).
- None of Grok 4.7’s coding comparisons are independently confirmed on both sides yet.
What changed from Grok 4.6 to Grok 4.7
The 5 benchmarks both models have been scored on, using the same variant each time. A raw Grok 4.7 gain is not a real gain until it clears that benchmark’s own noise band, so each row below carries the verdict and not just the arithmetic.
| Benchmark | Grok 4.6 | Grok 4.7 | Change | Verdict |
|---|---|---|---|---|
| HLE | 42.9 | 43.1 | +0.2 | Setup-dependent |
| Terminal-Bench 2.1 | 88.4 | 73.41 | -15.0 | Setup-dependent |
| DeepSWE | 67 | 71 | +4.0 | Unverified |
| LiveBench | 78 | 77.4 | -0.6 | Tie |
| AnalystAgent | 41.25 | 52.5 | +11.3 | Setup-dependent |
Grok 4.7 API pricing
$2.00 in / $6.00 out per 1M tokens — official pricing source
What Grok 4.7 costs per job
| Workload | Tokens in / out | Cost |
|---|---|---|
| One long chat turn | 100K / 10K | $0.260 |
| A codebase review | 1,000K / 100K | $2.60 |
| A day of agent work | 10,000K / 1,000K | $26.00 |
Computed from Grok 4.7’s list rates above — cache discounts and batch tiers are not applied.
Grok 4.7 is one of 2 xAI (SpaceXAI) models tracked on this site, at these official list prices.
| Model | In / 1M | Out / 1M |
|---|---|---|
| Grok 4.7 | $2.00 | $6.00 |
| Grok 4.6(previous version) | $2.00 | $6.00 |
Grok 4.7 benchmark scores
| Benchmark | Score |
|---|---|
| HLE(no tools)[1] Reasoning · ±2 is noise | |
| LiveBench[2] Composite score across 7 domains · ±2.7 is noise | |
| DeepSWE[3] Long-horizon coding · ±9.5 is noise | |
| Terminal-Bench 2.1[4] Terminal ops · ±10.6 is noise | |
| Terminal-Bench 4.0(xhigh)[5] Terminal ops · ±12.4 is noise | |
| AnalystAgent[6] Spreadsheet & document analysis · ±11.2 is noise |
Who ran these numbers: 5 of 6 independent — artificialanalysis.ai (2), vals.ai (2), livebench.ai (1); vendor self-reported (1).
- HLE: AA's own run at xhigh effort (Grok 4.6's tracked score used high effort) — text-only 2,158-question subset, part of Intelligence Index v4.3.2. Index is now at v5 (2026-10-28); that HLE cell remains a tracked single-benchmark score, not an Index entry.
- LiveBench: Board row 'Grok 4.7 xHigh' on the same 2026-06-25 LiveBench release Grok 4.6 sits on; only the xHigh tier is populated. Grok 4.6 scores slightly higher overall on this table (78.0 vs 77.4).
- DeepSWE: xAI's own model card figure (DeepSWE v1.1, high effort). The independent deepswe.datacurve.ai board (checked live 2026-09-23, 28 tracked models) does not list this model yet — only grok-4.6 and grok-4.5 appear. xAI's table marks this number with an asterisk defined as 'a high-effort score', below the column's xHigh tier — the effort field records it.
- Terminal-Bench 2.1: vals.ai's own Terminus 2 run, board row 'Grok 4.7' (73.41 ± 1.50, pass@1, reasoning effort xhigh), from its archived Terminal-Bench 2.1 table.
- Terminal-Bench 4.0: vals.ai board row — mini-swe-agent harness (single bash tool), pass@1 averaged over 3 full passes, raw value 28.788 at effort xhigh. Artificial Analysis same-conditions run (Xhigh): 25.76. Official board (grant-funded, Grok Build, xhigh effort, 5-trial pass@1): 37.58 ± 3.54 CI — 8.8 points above this run.
- AnalystAgent: AA's own run, board row 'Grok 4.7 (Xhigh)': analystAgent 0.525 = 52.5 pass^5. pass@1 is 63.5. Read from the Grok 4.7 (Xhigh) page on 2026-10-15.
Notes on the record
Pricing follows the identical two-tier scheme as Grok 4.6, confirmed on docs.x.ai/docs/models (checked 2026-09-23): $2.00 input / $6.00 output per million tokens under 200,000 prompt tokens, rising to $4.00/$12.00 — the full request billed at the higher rate — once a prompt reaches 200,000. Cached input follows the same split, $0.50 rising to $1.00. A "Fast" variant bills at twice the standard rates; xAI's developer docs say it is available only in Cursor and Grok Build, not the public API.
xAI's developer docs list no text output limit. Knowledge cutoff is May 2026 per the docs, while the model card gives a June 2026 pretraining cutoff — both official, recorded unresolved. Parameter count is undisclosed in xAI's model card; widely circulated third-party reports of 2.1 trillion parameters trace to no xAI source found and are not used here. Input: text and images. Output: text only.
Independent coverage came slowly. Grok 4.6's first eight independent scores arrived 5 to 12 days after its 2026-08-12 release, not at launch (a ninth, AA-AnalystAgent, came when this site added that board on 2026-09-29). Grok 4.7's record two days in was Humanity's Last Exam (Artificial Analysis, text-only, xhigh effort, a different tier from Grok 4.6's tracked high-effort score) and LiveBench. vals.ai's Terminal-Bench 2.1 run (73.41, xhigh) was added 2026-10-01; Grok 4.6's row is Artificial Analysis's, so the pair reads setup-dependent. xAI's self-reported DeepSWE v1.1 score is below; its Terminal-Bench 4.0 score of 38.0% is a different benchmark version and is not recorded.
xAI's own materials never say Grok 4.7 supersedes or replaces Grok 4.6 — Grok 4.6 stays listed on the same pricing table, not flagged deprecated. "xAI (SpaceXAI)" reflects the same July 2026 corporate rebrand already explained on the Grok 4.6 page.
Compare with
FAQ
Has Grok 4.7 been independently benchmarked?
Partly. Artificial Analysis has an independent Humanity's Last Exam score (43.1%, text-only, xhigh effort) and LiveBench has an independent board row (77.4 overall, xHigh), both from launch week. vals.ai's archived Terminal-Bench 2.1 board adds a third, 73.41 at xhigh, recorded 2026-10-01, and Terminal-Bench 4.0 (28.79, xhigh) a fourth, added 2026-10-08. AA-AnalystAgent reads 52.5 pass^5 at xhigh (Artificial Analysis, checked 2026-10-15). GPQA Diamond, SWE-bench Verified, LiveCodeBench, ARC-AGI-2 and the independent DeepSWE leaderboard had no Grok 4.7 row when checked; deepswe.datacurve.ai and arcprize.org were checked directly on 2026-09-23.
Does Grok 4.7's price change with longer prompts?
Yes, the same way Grok 4.6's does. The $2 input / $6 output per-million-token rate applies under 200,000 prompt tokens; at 200,000 and above, xAI bills the whole request at $4 input / $12 output, per docs.x.ai/docs/models (checked 2026-09-23). Cached input doubles the same way, $0.50 to $1.00.
How many parameters does Grok 4.7 have?
xAI has not disclosed a parameter count in its official model card or documentation. Third-party reports circulating online cite 2.1 trillion parameters, but no xAI source states this figure, so it isn't used on this page.
Does Grok 4.7 replace Grok 4.6?
Not according to xAI's own materials. Grok 4.6 remains listed on the same pricing page, with no deprecation notice, and xAI's announcement and model card describe Grok 4.7 as extending Grok 4.6's capability rather than retiring it — "served at the same price and speed as Grok 4.6," per xAI's own launch post.
Is Grok 4.7's Terminal-Bench score comparable to other models on this site?
xAI's own figure is not: its self-reported 38.0% is on Terminal-Bench 4.0, a different, newer benchmark version from the Terminal-Bench 2.1 this site tracks. The Terminal-Bench 2.1 score on this page is vals.ai's independent run, 73.41 at xhigh effort. It compares directly with other vals.ai rows, such as Grok 4.6's 78.28 on the same board, a tie; against Artificial Analysis or tbench.ai rows it reads setup-dependent.
What is Grok 4.7's knowledge cutoff?
May 2026, again at month precision. The context window stays 500,000 tokens, and prompts at or above 200,000 tokens bill at the higher long-context tier — same structure as the pricing question above describes.
Further reading
- Models with 10M token context windows 2026 — Grok 4.7 is one of the 47 models it compares.
Benchmark guides
- Humanity's Last Exam — what Grok 4.7’s reasoning score on it does and doesn’t prove.
- LiveBench — what Grok 4.7’s composite score on it does and doesn’t prove.