Elon Musk’s SpaceXAI to Release Grok 4.5 as GPT-5.6 and Fable 5 Intensify AI Race

Elon Musk's SpaceXAI released Grok 4.5 on Wednesday, its first public model since the SpaceX-xAI merger closed in February and SpaceX's pending $60 billion deal to acquire Cursor. The model targets coders, engineers, and knowledge workers, priced at $2 per million input tokens and $6 per million output—far below Claude Opus 4.8 at $5 input and $25 output, and GPT 5.6 Sol at $5 input and $30 output. Musk called it 'roughly comparable to Opus 4.7, but much faster,' framing a deliberate tradeoff of speed and cost over raw capability. Benchmarks were mixed: Grok 4.5 scored 53% on DeepSWE 1.1 versus 59% for Opus 4.8 and 70% for Claude Fable 5, but beat GPT 5.5 on SWE Bench Pro with 64.7%.

Background

The release follows SpaceX's February merger with xAI and its pending $60 billion acquisition of Cursor, the AI coding platform. Grok 4.5 was trained in collaboration with Cursor on tens of thousands of Nvidia GB300 GPUs inside Colossus, the Memphis supercomputer with total capacity across more than 200,000 GPUs. According to the source, the labs whose models sit above Grok 4.5 on benchmarks don't own anything close to that hardware. SpaceXAI has consistently turned up with enormous compute and third-place scores, a pattern that has followed Grok across multiple releases. What changed with Grok 4.5 is the pricing and the training signal. The model was trained on developer session data from Cursor, including debugging traces and real code edits rather than static repositories. Musk admitted in court that xAI's training practices have drawn scrutiny before; this time the pipeline runs through a platform SpaceX is in the process of buying outright. The launch also coincides with new top-tier models from OpenAI (GPT 5.6 Sol) and Anthropic (Claude Fable 5), intensifying the AI race.

What happened

SpaceXAI released Grok 4.5 on Wednesday, according to Decrypt. It is the first public model since the SpaceX-xAI merger closed in February and SpaceX's pending $60 billion deal to acquire Cursor. The company pitches the model as cheap for a western model: $2 per million input tokens and $6 per million output, compared with Claude Opus 4.8 at $5 input and $25 output, and GPT 5.6 Sol at $5 input and $30 output. Musk posted on X that Grok 4.5 is 'roughly comparable to Opus 4.7, but much faster,' framing speed and cost over raw capability, with engineers at Tesla and SpaceX as proof of real-world utility. SpaceXAI published four benchmark results. On DeepSWE 1.1, which measures how reliably an AI closes real software bugs, Grok 4.5 scored 53%, behind Claude Opus 4.8 at 59% and GPT 5.5 at 67%, with Claude Fable 5 topping the chart at 70%. On SWE Bench Pro, Grok 4.5 posted 64.7%, beating GPT 5.5's 58.6%, while Opus 4.8 led at 69.2% and Fable 5 at 80.4%. The benchmarks compare against GPT 5.5, not GPT 5.6, because the latter launched Wednesday hours after Grok 4.5's announcement. Grok 4.5 was trained with Cursor AI on tens of thousands of Nvidia GB300 GPUs inside Colossus, which has total capacity across more than 200,000 GPUs. On SWE Bench Pro tasks, Grok 4.5 used an average of 15,954 output tokens per job, while Opus 4.8 burned 67,020 tokens, a 4.2x gap. The model runs at 80 tokens per second and offers half a million tokens of context. It is available via API, on Hermes and Grok build; European users will have to wait.

Market impact

The pricing figures are the clearest market signal in the source. Grok 4.5 costs $2 per million input tokens and $6 per million output. Claude Opus 4.8, Anthropic's primary flagship, runs $5 input and $25 output. GPT 5.6 Sol, OpenAI's new top-tier model that also launched Wednesday, is priced at $5 input and $30 output. That makes Grok 4.5 roughly 60% less per input token than Opus 4.8, a figure the source explicitly cites. On efficiency, Grok 4.5 used an average of 15,954 output tokens to complete each SWE Bench Pro task, while Opus 4.8 burned through 67,020 tokens for the same work, a 4.2x gap. For teams running AI at volume, the source says that difference compounds into real savings on top of the already-lower price per token. Even with Grok 4.5 scoring lower on quality benchmarks, cheaper tokens and more efficient usage allow for more iterations without spending so much. The model runs at 80 tokens per second, which the source calls fast-model territory. No trading volumes, token flows, or market capitalizations are given in the source. The competitive landscape includes Claude Fable 5, Anthropic's top of the line offering, which leads every category SpaceXAI chose to publish.

Why it matters

The source frames Grok 4.5 as a deliberate tradeoff: speed and cost over raw capability. For developers running high-volume coding tasks, the math works out to roughly Opus 4.7 capability at 60% less per input token. That could appeal to teams that prioritize iteration speed and budget over frontier performance. The model was trained on developer session data from Cursor, including debugging traces and real code edits rather than static repositories, a pipeline that runs through a platform SpaceX is in the process of buying outright. Musk admitted in court that xAI's training practices have drawn scrutiny before, so the Cursor data pipeline may attract further attention. The release also highlights a pattern: SpaceXAI has consistently turned up with enormous compute and third-place scores. What changed with Grok 4.5 is the pricing and the training signal. The better argument isn't raw performance; it's efficiency math. For anyone chasing the frontier, Claude Fable 5 leads every category SpaceXAI chose to publish. The source's quick test using Grok build on Hermes was underwhelming for creative writing and acceptable on a simple coding task. European users will have to wait before using it.