Verified · Jul 9, 2026

xAI 7/8 Grok 4.5 launch: DeepSWE 62% / SWE Marathon 29% / ~4.2x fewer output tokens than Opus 4.8, default in Cursor across all plans on launch day

2 sources

xAI's July 8, 2026 launch post introduces Grok 4.5 as SpaceXAI's newest flagship, built for coding, agentic tasks, and knowledge work; the post says the model was co-trained with Cursor on 'tens of thousands of NVIDIA GB300 GPUs.' Self-reported benchmarks: DeepSWE 1.0 62.0% (vs Fable max 66.1%, GPT 5.5 xhigh 64.31%); SWE Marathon 29.0% resolution rate (highest reported); Terminal Bench 2.1 83.3%; SWE Bench Pro 64.7%; on SWE Bench Pro, ~4.2x fewer output tokens than Opus 4.8 max (15,954 vs 67,020). Served at ~80 TPS, priced $2 / $6 per million input / output tokens; available in Grok Build, Cursor (all plans), and the SpaceXAI API at launch; not yet available in the EU in any SpaceXAI product or API console (mid-July ETA).

Why now

xAI's 7/8 launch is the strongest single-flagship event of early July — it ships self-reported benchmarks vs Opus / GPT / Fable, a concrete token-efficiency number, and 'default in Cursor across all plans on launch day' as a single bundle. English-speaking coding audiences get a model they can try in Cursor the same day.

Why it is worth publishing

Huge demo surface: Cursor users see Grok 4.5 in the model picker on launch day; you can record a single-prompt SWE task switching between Grok 4.5 / Opus 4.8 / GPT 5.5 to show the ~4.2x output-token delta.

Evidence basis

xAI flagship launch + default-in-Cursor across all plans on launch day + multiple self-reported benchmarks at once — heat is medium-to-high as a single-flagship event, but driven by official self-reporting rather than independent reproduction.

“Grok 4.5 just landed as the default in Cursor across every plan — it claims 64.7% on SWE Bench Pro but uses ~4.2x fewer output tokens than Opus 4.8.”

Angle

Frame the Grok 4.5 7/8 launch as the strongest single-flagship event of early July — bundle self-reported benchmarks vs Opus / GPT / Fable, the concrete token-efficiency number (~4.2x), and 'default in Cursor across all plans on launch day' into one piece rather than reading any single benchmark in isolation.

Format

Short talking-head video

Demo idea

Record a 30-second Cursor screen capture: run the same SWE task in Grok 4.5 / Opus 4.8 / GPT 5.5 by switching models in the picker, then overlay the output-token counts to show the ~4.2x delta; close with 'EU ships mid-July, Cursor default across all plans today.'

Platform notes

xAI's benchmarks are self-reported (medium risk) — don't paraphrase the launch as third-party validation; the exact GB300 cluster size isn't stated (medium risk); the model isn't yet available in the EU (medium-low risk) — flag for EU audiences.

Usable claims

xAI's July 8, 2026 launch post introduces Grok 4.5 as SpaceXAI's newest flagship, built for coding, agentic tasks, and knowledge work; the post states the model was trained with Cursor on 'tens of thousands of NVIDIA GB300 GPUs' and reports DeepSWE 1.0 62.0% (vs. Fable max 66.1%, GPT 5.5 xhigh 64.31%), SWE Marathon 29.0% resolution rate (highest reported), Terminal Bench 2.1 83.3%, SWE Bench Pro 64.7% resolve rate, and ~4.2x fewer output tokens than Opus 4.8 (max) on SWE Bench Pro (15,954 vs 67,020); served at ~80 TPS, priced at $2/$6 per million input/output tokens; already available in Grok Build, Cursor (all plans), and the SpaceXAI API; not yet available in the EU in any SpaceXAI product or API console (mid-July ETA).
xAI's July 8, 2026 launch post frames Grok 4.5 as a co-training effort with Cursor on 'tens of thousands of NVIDIA GB300 GPUs' and ships Grok 4.5 as a default option in Cursor across all plans at launch.

Evidence pipeline

From the news

Breakdown

Reading any single Grok 4.5 fact point (DeepSWE 62% / SWE Marathon 29% / ~4.2x fewer output tokens than Opus 4.8 / default in Cursor across all plans) in isolation turns into 'reading benchmarks.' This piece explains how to use the 'strongest single-flagship event of early July' frame — bundle self-reported benchmarks vs Opus / GPT / Fable + the concrete token-efficiency number + 'default in Cursor across all plans on launch day' into a single piece so creators can produce 'a model you can try in Cursor today' content rather than benchmark reading.

Sources

Risks

Pin links to each source; quote only what the captured summary states; do not paraphrase specific benchmark numbers, performance metrics, paper claims, or architectural details beyond what is stated.
Pin the link to xAI's Grok 4.5 launch post; quote only what the post states; do not paraphrase product positioning as third-party validation; do not generalize cluster-framing to other xAI models; flag that the launch is an xAI self-reported announcement and that the model is not yet available in the EU.

Demo ideas

In Cursor, run the same SWE task in Grok 4.5 / Opus 4.8 / GPT 5.5 by switching models and overlay output-token counts, latency, and resolve rate side-by-side.
Build a four-model SWE benchmark comparison card: Grok 4.5 vs Opus 4.8 vs GPT 5.5 vs Fable max, calling out Grok 4.5's token-efficiency advantage.
Run the same agentic task on Grok Build and the SpaceXAI API to show the ~80 TPS serving experience.