Grok 4.7 Launches: Big Benchmark Gains but Doubled Token Use Sparks Cost Debate

Grok 4.7 was announced by Elon Musk on 09-22 and immediately put through third-party benchmarks, led by Artificial Analysis (AA). The headline conclusion: substantial capability gains paired with sharply higher token usage. AA scored it 46 on the Intelligence Index (2 points above Grok 4.6), putting xAI in the top four labs for the first time; on the multi-hour office benchmark AA-Briefcase it took second place with 58%, just 1 point behind Claude Fable 5.1 Max (59%) and ahead of GPT-6 Astra. But at roughly 81k output tokens per task—more than double its predecessor—its per-task cost exceeded Astra in AA's comparison, fueling a community debate over value.

Confirmed

• AA evaluation: Intelligence Index 46 (Grok 4.6 at 44); hallucination rate down from 34% to 29%; output speed 188 tokens/s, 7.1 minutes average per Intelligence Index task
• AA-Briefcase: 1657 Elo, +111 over Grok 4.6 (high), behind only Claude models; analysis-quality Elo at 1994, +88 over GPT-6 Astra; GDPval-AA ahead of Astra by +153 Elo
• Win-rate: 58%, second only to Claude Fable 5.1 Max at 59%
• AA stated that in the AA-Briefcase due-diligence scenario, Grok 4.7's per-task cost is about half of Opus 5's
• Coding: CursorBench 4.0 at 46.3%, nearly matching Opus 5 Max's 46.6%; CursorBench up from 40.4%, EEBench from 53% to 64%; Coding Agent Index rose from 47 to 56 with the Grok Build agent
• Cost: 81k output tokens per task for xhigh, over double Grok 4.6 (xhigh); AA's comparison put per-task cost above GPT-6 Astra
• Showcase cases: independently running a step-by-step valuation chain and challenging the counterparty's estimate; analyzing three full years of financials to expose FX-masked growth, where 4.6 used only the latest year
• FinanceYF5 compared open-world city game demos by 4.7 vs 4.6, finding 4.7 worked longer, verified more, and produced better output at unchanged pricing

Unconfirmed

• XFreeze's personal benchmark: on an identical coding task, Grok 4.7 xHigh matched Opus 5 Max at under half the cost with fewer tokens and steps—a personal measurement that doesn't contradict AA's Astra comparison
• One blogger's same-prompt test at max reasoning: Grok 4.7 took 32 minutes and $8.14 versus free SWE 2 at 15 minutes and $0
• ChrisGPT claimed Grok 4.8 arrives by end of November and that Grok 4.7's per-task cost exceeds GPT's—unverified leaks

Why it matters

• Users including Angaisb and Theo (repost) noted Grok 4.7 uses nearly 3x Astra's tokens per task and 2.25x Grok 4.6's, meaning real costs under "same price, same speed" marketing may be higher; Angaisb argued OpenAI is the only lab serious about token efficiency, and Theo's repost said Grok 4.7 is no longer cheap, costing more than Astra in AA's comparison
• Critics (relayed by ivanbezdomny) questioned the value of a 2-point gain at 2.5x token consumption
• For heavy API users, the model now matches frontier capabilities at higher unit cost, requiring workload-specific recalculations; meanwhile its deeper multi-hour agent analysis is a clear capability gain for professional work

2026-09-22 ~ 2026-09-22 · 25 related posts

Full story(9 episodes)→

• Episode 1: Grok 4.7 Briefly Surfaces in OpenCode Zen Directory, Fueling Launch Rumors(2026-09-21, 5 posts)
• Episode 2: xAI launches Grok 4.7 with major benchmark gains at unchanged price and speed(2026-09-21, 59 posts)
• Episode 3: Grok 4.7 Day-One Hands-On: Strong Word-of-Mouth for Autonomous Coding(2026-09-22, 17 posts)
• Episode 4: Grok 4.7 Launches: Big Benchmark Gains but Doubled Token Use Sparks Cost Debate(2026-09-22, 25 posts)
• Episode 5: Grok 4.7 Reportedly Closes Gap With Top Frontier Models at Lower Cost(2026-09-22, 2 posts)
• Episode 6: Open-Source Grok 4.7 Editor Extension Passes 140K Installs(2026-09-22, 3 posts)
• Episode 7: Bug Hunt Bench: GPT-6 Astra Tops 105-Bug Fixing Test, MiMo Stuns on Value(2026-09-22, 18 posts)
• Episode 8: Grok 4.7 Meets Xiaomi MiMo-V2.6 in a Same-Night Price War(2026-09-22, 5 posts)
• Episode 9: xAI Ships Three Grok Frontier Models in Nine Weeks(2026-09-22, 2 posts)

• Grok 4.7 Reportedly Uses Nearly 3x More Tokens per Task Than Astra — Angaisb_ · 2026-09-22
• [source] Grok 4.7 scores 46 on AA Intelligence Index, enters top 4 labs but burns 81k tokens per task — ArtificialAnlys · 2026-09-22
• Grok 4.7's AA-Briefcase analytical quality Elo jumps to 1994 from 1690 — ArtificialAnlys · 2026-09-22
• Grok 4.7 + Grok Build jumps to 56 on AA Coding Agent Index, up from 47 — ArtificialAnlys · 2026-09-22
• Grok 4.7 jumps on coding index but burns 81k tokens per task, 2x its predecessor — ArtificialAnlys · 2026-09-22
• [source] Grok 4.7's 81k output tokens per task more than double Grok 4.6's 36k — ArtificialAnlys · 2026-09-22
• Grok 4.7 measured at ~188 tokens/second, ~7.1 minutes per Intelligence Index task — ArtificialAnlys · 2026-09-22
• Same-prompt test: Grok 4.7 takes 32 min at $8.14 while free SWE 2 finishes in 15 min — iamfakhrealam · 2026-09-22
• Grok 4.7 cuts hallucination rate to 29% from 34% on AA-Omniscience — ArtificialAnlys · 2026-09-22
• Grok 4.7 matches Opus 5 Max on coding at less than half the cost — XFreeze · 2026-09-22
• Grok 4.7 xHigh hits 46.3% on CursorBench 4.0, matching Opus 5 Max at half the cost — XFreeze · 2026-09-22
• Grok 4.7 xHigh hits 58% on AA-Briefcase, just 1 point behind Claude Fable 5.1 Max — XFreeze · 2026-09-22
• Unverified: Grok 4.7 out now, costs more per task than GPT, says leaker — ChrisGPT · 2026-09-22
• Grok 4.7 Slammed in Benchmarks: 2.5x Tokens for Just 2 Points More — ivan_bezdomny · 2026-09-22
• [source] Grok 4.7 ranks just behind Anthropic's Opus 5 on AA-Briefcase at ~50% of the cost per task — ArtificialAnlys · 2026-09-22
• AA example: Grok 4.7 independently runs valuation chain and flags divergence from deal partner — ArtificialAnlys · 2026-09-22
• Grok 4.7 example: three-year financial analysis exposes currency-masked growth stall — ArtificialAnlys · 2026-09-22
• Grok 4.7 jumps to 46.3% on CursorBench and 64% on EEBench, keeping the same $2/$6 per million token pricing — FinanceYF5 · 2026-09-22
• Grok 4.7 looks pricier than before, now costlier than Astra on Artificial Analysis — steipete · 2026-09-22
• Grok 4.7 Token Efficiency Falls 30-80% Short of Claims, Real-World Costs 2x Grok 4.6 — flowersslop · 2026-09-22
• Grok 4.7 beats GPT-6 Astra on professional work, leading by 153 Elo on GDPval-AA — XFreeze · 2026-09-22
• Grok 4.7 costs $2.73 per task vs $1.86 for a +2 point AA index gain — koltregaskes · 2026-09-22
• Grok 4.7 lands: near Opus 5 on AA-Briefcase at ~50% cost per task — NicoVerderosa · 2026-09-22

2 near-duplicate retellings: Angaisb_ · CurieuxExplorer