SpaceXAI has launched Grok 4.5, pitching the model to developers and enterprises trying to control the rising cost of AI-assisted software development.
In a statement, the company said the model is priced at $2 per million input tokens and $6 per million output tokens. It said the model is built for coding and agentic work, runs at 80 tokens per second, and uses fewer tokens than comparable models on some software engineering tasks.
Grok 4.5 is available through the SpaceXAI console and Grok Build. It is also available in Cursor, the AI coding tool made by Anysphere, giving SpaceXAI a route into a development environment already used by programmers rather than only competing through an API. SpaceXAI said EU availability is expected in mid-July.
In June, SpaceX, which owns SpaceXAI, said it was buying Anysphere, the startup behind Cursor, in a deal aimed at strengthening its position in enterprise AI tools. In a separate statement, Cursor said that Grok 4.5 was trained jointly with SpaceXAI and used trillions of tokens of Cursor data, including user interactions with codebases and software tools.
The launch addresses a growing realization among enterprise engineering teams that AI coding agents can become expensive once they move beyond simple prompts.
“Enterprises are hitting a wall with AI ROI,” said Neil Shah, vice president for research at Counterpoint Research. “The massive token consumption required by autonomous agents and coding is causing bill shocks, turning AI adoption into an expensive, one-way street.”
On Artificial Analysis’ Coding Agent Index, Grok 4.5 in Grok Build finished below Fable 5 in Claude Code and roughly level with GPT-5.5 in Codex. It estimated Grok 4.5’s cost at $2.49 per task, compared with $5.07 for GPT-5.5 in Codex and $11.80 for Fable 5 in Claude Code.
The figures give SpaceXAI a useful proof point, though analysts said companies will still need to test the model on their own codebases before relying on it widely.
“It is too early to say if Grok 4.5 is a game changer,” said Anand Joshi, managing director of market research firm JP Data. “The benchmarks are impressive, and the low token usage will be attractive to enterprises. The developer community will give a verdict in time if the coding output is superior to the competition.”
“Grok 4.5’s pricing is notable because it lowers the economics of running agentic coding workloads, but enterprise buyers should focus on cost per successful outcome rather than cost per token,” said Biswajeet Mahapatra, principal analyst at Forrester.
A cheaper model can still cost more in practice if it needs repeated attempts to produce working code, Mahapatra said. Enterprises should look at the full cost of a coding workflow, including developer review effort and whether the final output is usable, he said.
A bigger concern, according to Lian Jye Su, chief analyst at Omdia, is that token use has become too easy a proxy for value.
“We are living in the era where token consumption is seen as the ultimate value creation but the true value still lies in actual job completion,” Su said. “To most enterprises, the cost per job done remains the best approach to assess agent effectiveness.”
That makes Grok 4.5 less a simple pricing story than a test of whether SpaceXAI can lower the actual cost of AI-assisted development in real engineering environments, where corporate codebases often expose weaknesses that public benchmarks may miss.
Mahapatra said tests such as SWE-Bench Pro, DeepSWE, and Terminal Bench can offer early signals, but enterprises should also compare Grok 4.5 with other models on their own repositories before adopting it more widely. Su said A/B testing in real development environments, combined with cost monitoring over time, would give enterprises a clearer view of token efficiency and output quality.
Grok 4.5 is unlikely to displace broader enterprise AI platforms on price alone. Its more realistic near-term role is in software engineering workflows, particularly at companies already using more than one model and trying to route work based on cost, speed, and accuracy.
Cursor and Grok Build users are among the most likely to find this model useful, according to Su. Mahapatra said Grok 4.5 could become a primary coding assistant for some teams, especially where software engineering is the main workload, but larger enterprises are more likely to test it as part of a mixed-model strategy.
Shah said that the shift is already underway as enterprises become more cautious about relying on a single AI provider. High-risk or more complex tasks may still go to models such as Claude, he said, while Grok 4.5 could appeal to high-volume developer workflows and repetitive agentic tasks if its accuracy proves close enough to rival systems.
Cursor could give SpaceXAI another advantage, Shah added. By training with developer interaction data from Cursor, Grok 4.5 could benefit from a feedback loop based on how programmers actually write, review, and debug code.
The article originally appeared on InfoWorld.