Grok 4.7: From Leak to Launch

Episode 1 · Grok 4.7 briefly surfaces on OpenCode Zen, launch rumored imminent (2026-09-21, 5 posts)

On September 21, xAI's next-generation model Grok 4.7 appeared to be nearing release: several X users spotted grok-4.7 briefly listed in the OpenCode Zen model directory before it was removed, widely read as a sign that a launch was imminent. That same day, another user claimed Grok 4.7 would drop that afternoon, sharing specific specs and pricing details. xAI has not officially confirmed anything so far.

Confirmed

• grok-4.7 did appear in the OpenCode Zen model directory and was subsequently removed. @kimmonismus, @CurieuxExplorer, and @socialwithaayan all reported the sighting; @CurieuxExplorer joked that if it's on the menu, it's already in the kitchen.
• xAI has made no official confirmation about the Grok 4.7 release as of now.

Unconfirmed

• @ns123abc posted that Grok 4.7 would launch early that afternoon, with screenshots as evidence, but there is no official backing.
• @realsohamparekh claimed Grok 4.7 would launch today with a 500K tokens context window, multimodal support, and pricing similar to Grok 4.6. The claim comes from a personal account and has not been verified by xAI.
• Both @kimmonismus and @socialwithaayan predicted this would be a "big model release week": after xAI, OpenAI (Dev Day) and Anthropic could make moves in quick succession — though this is personal speculation.

Why it matters

• A model name briefly appearing in the OpenCode Zen directory is typically seen as a precursor to an official rollout, so the community views this incident as a strong signal that xAI's new model is about to launch.
• If the rumors of a 500K context window, multimodal capabilities, and pricing on par with Grok 4.6 hold true, Grok 4.7 would compete directly with rivals on long-context performance and value for money — but everything hinges on xAI's official announcement.

• Unverified claim: Grok 4.7 launching today with 500K context, multimodal, 4.6-level pricing — realsohamparekh · 2026-09-21
• Grok 4.7 rumored to drop this afternoon, unconfirmed — ns123abc · 2026-09-21
• Grok 4.7 briefly appears on OpenCode Zen before deletion, hinting at imminent release — kimmonismus · 2026-09-21
• Grok 4.7 briefly spotted on opencode zen, hinting at imminent xAI release — socialwithaayan · 2026-09-21
• Grok 4.7 Spotted in OpenCode Zen Model Catalogue, Release Imminent — CurieuxExplorer · 2026-09-21

Episode 2 · Grok 4.7 rolls out at same price with big gains on long-horizon tasks (2026-09-21, 5 posts)

xAI has released Grok 4.7, with multiple community users confirming the new model is live and reportedly appearing in the xAI API, corroborating the launch news. Taken together, the leaks suggest the new version delivers significant benchmark and capability improvements at unchanged pricing—the most notable story of this release cycle.

Confirmed

• Grok 4.7 has shipped and is available: users including @DanielLockyer confirmed the new version is live, and @iamfakhrealam reports, citing LuminaBench, that Grok 4.7 has appeared in the xAI API.
• Pricing stays flat versus 4.6: $2/M input, $6/M output, plus a double-speed tier at twice the price (via @markk).
• Upgrade direction: a larger base model with longer RL training; officials say the new model can persist longer on hard multi-hour tasks and check its results more carefully (via @nimaowji, @markk).
• Benchmark performance: Terminal-Bench improved from 20.3% to 38.0% (via @markk, noted as unverified in the post).

Not Yet Confirmed

• Most posts are leaks from third-party accounts, and early posts (@markk, @koltregaskes) explicitly noted they were unverified by official xAI channels; later community confirmations and the API sighting boost credibility, but a full official statement from xAI and detailed benchmark data are still pending.

Why It Matters

• Faster performance at the same price, plus a near-doubling on Terminal-Bench, directly improves Grok's cost-performance competitiveness in agent and terminal task scenarios.
• The emphasis on multi-hour long-horizon tasks and self-checking signals that xAI is positioning agent scenarios as its main competitive front.

• Grok 4.7 reportedly released, details still unconfirmed — koltregaskes · 2026-09-21
• Grok 4.7 spotted in xAI API ahead of confirmation — iamfakhrealam · 2026-09-22
• Grok 4.7 is live, community confirms rollout — DanielLockyer · 2026-09-22
• Unverified: Grok 4.7 Leak Claims Same Price, Terminal-Bench Jump From 20.3% to 38.0% — mark_k · 2026-09-22
• Grok 4.7 reportedly out: same price as 4.6, bigger base, longer training on hard multi-hour tasks — nima_owji · 2026-09-22

Episode 3 · xAI Launches Grok 4.7, Its Strongest Coding Model Yet (2026-09-22, 26 posts)

xAI (SpaceXAI) officially released Grok 4.7 on September 22, positioning it as its most powerful model for coding and knowledge work. The company claims it runs twice as fast as comparable models at half the price—actual pricing stays level with Grok 4.6 at $2/million input tokens and $6/million output tokens. The notable jump in benchmark scores is the highlight of this release.

Confirmed

• Grok 4.7 is built on a new, larger base model with extended reinforcement learning for multi-step tasks lasting hours; it is better at self-verification and long-context management, and natively understands the Grok Bot system
• CursorBench score reaches 46.3%, surpassing GPT-5.6 Sol's 41.7%; Terminal-Bench leaps from the previous generation's 20.3% to 38%; Harvey legal benchmarks also improved substantially
• Pricing is unchanged from Grok 4.6: $2/million input tokens, $6/million output tokens, which the company says is a fraction of what comparable competitors charge

Not Yet Confirmed

• Third-party commentary from blogger XFreeze claims Grok 4.7 shows a massive leap on multi-hour office tasks (AA Briefcase), surpassing GPT-6 Astra and approaching Fable 5.1—this is personal interpretation; refer to official benchmark tables for exact scores

Why It Matters

• Major benchmark gains at the same price and speed, combined with lower pricing than competitors, position Grok 4.7 to shake up the price-performance landscape for coding and knowledge-work models; the strengthened ability to handle long multi-step tasks with self-verification also targets real needs of agentic office workflows

• Grok 4.7 launches at $2/$6 per M tokens, hits 46.3% on CursorBench 4.0 — Scobleizer · 2026-09-22
• Grok 4.7 out: Terminal-Bench jumps 20.3% to 38%, beats GPT-5.6 Sol on CursorBench — thesaraharminta · 2026-09-22
• xAI ships Grok 4.7, update via single terminal command — tetsuoai · 2026-09-22
• xAI launches Grok 4.7: frontier coding model at $2/$6 per million tokens — tetsuoai · 2026-09-22
• Grok 4.7 launches: 46.3% on CursorBench, Terminal-Bench nearly doubles, same $2/$6 pricing — tetsuoai · 2026-09-22
• SpaceXAI Unveils Grok 4.7, Claiming Twice the Speed at Half the Price — Polymarket · 2026-09-22
• Grok 4.7 Released: Terminal-Bench Score Nearly Doubles in Push for Multi-Hour Tasks — Scobleizer · 2026-09-22
• Grok 4.7 arrives with notable gains over 4.6 at same price and speed — SpaceXAI · 2026-09-22
• Grok 4.7 works longer on hard tasks, ships strongest safeguards yet — SpaceXAI · 2026-09-22
• SpaceXAI Demos Grok 4.7 vs 4.6 Building the Same Open-World City Game — SpaceXAI · 2026-09-22
• xAI Launches Grok 4.7: Same Price, Frontier Price-Performance With 46.3% on CursorBench — SpaceXAI · 2026-09-22
• Grok 4.7 Shows Huge Jump on Multi-Hour Office Work, Nearly Matching Fable 5.1 — XFreeze · 2026-09-22
• Grok 4.7 lands in Grok Build: 4 reasoning tiers, Terminal-Bench jumps 20% to 38% — XFreeze · 2026-09-22
• Grok 4.7 ships at same price and speed, touted as a notable jump over 4.6 — elonmusk · 2026-09-22
• Grok 4.7 ships with notable gains in software engineering at same price as 4.6 — EricBuess · 2026-09-22
• Grok 4.7 ships with big agentic gains: Terminal-Bench jumps from 20.3% to 38.0% at unchanged pricing — omnivaughn · 2026-09-22
• Grok 4.7 released: notable improvement over Grok 4.6 at same price and speed — reach_vb · 2026-09-22
• Grok 4.7 launches with notable improvement over Grok 4.6 at the same price and speed — bookwormengr · 2026-09-22
• Musk announces Grok 4.7, touting intelligence, speed, low cost and strongest safeguards yet — tetsuoai · 2026-09-22
• Grok 4.7 ships, reportedly Pareto-optimal on CursorBench 4.0 — brandon_galang · 2026-09-22

6 more related posts →