Grok 4.5
SpaceXAI's new workhorse, released 9 July 2026 — pitched by Musk as "an Opus-class model, but faster, more token-efficient and lower cost," at a $2/$6 rate card that undercuts every Western flagship.
Grok 4.5 specs at a glance
| Maker | SpaceXAI (formerly xAI) |
| Released | Announced 8 July 2026 · public 9 July 2026 |
| Model ID | grok-4.5 |
| Context window | 500,000 tokens |
| API pricing | $2 / $6 per 1M input / output tokens |
| Reasoning | Low, medium or high effort (default high) |
| Tools | Function calling, web search, X search, code execution |
| APIs | Responses API and Chat Completions (xAI API console) |
| Speed | ~80 output tokens/second (SpaceXAI figure) |
| Availability | GA — API console, Grok Build, Cursor, OpenRouter and other gateways · EU expected later in July 2026 |
| Best at | Coding and app-building, office and clerical work, research, routine knowledge work |
Calculate API costs for this model → Compare its monthly bill against 15 other rate cards.
The "Opus-class" claim
Elon Musk introduced Grok 4.5 as "an Opus-class model, but faster, more token-efficient and lower cost" — and separately as "roughly comparable to Opus 4.7, but much faster." SpaceXAI positions it not as a frontier-benchmark champion but as a workhorse: a model for the routine knowledge work the industry is trying to automate — coding and app-building, office and clerical tasks, research and writing.
The published numbers largely back the framing: on SpaceXAI's own benchmark chart, Grok 4.5 beats Claude Opus 4.8 on two of the four coding benchmarks shown and trails it on the other two — competitive with the top models, just short of best-in-class. Where it clearly wins is cost per solved task: a rate card 2.5-4x cheaper than Opus 4.8's, multiplied by far fewer tokens used per task.
Benchmarks
Figures below are from SpaceXAI's launch materials as reported by independent coverage. As always, vendor-published benchmarks favour the vendor's chosen suites.
| Benchmark | Grok 4.5 | Comparison (same chart) |
|---|---|---|
| SWE-bench Pro | 64.7% | Claude Fable 5: 80.4% · Claude Opus 4.8: 69.2% |
| Terminal-Bench 2.1 | 83.3% | Claude Fable 5: 84.3% · GPT-5.5: 83.4% |
| DeepSWE 1.1 | 53% | Claude Fable 5: 70% · GPT-5.5: 67% · Claude Opus 4.8: 59% |
| Token efficiency (SWE-bench Pro) | 15,954 avg output tokens/task | Claude Opus 4.8 (max): 67,020 — ~4.2x more |
The token-efficiency row is the real headline. SpaceXAI claims "twice greater token efficiency" than other leading models in general, and on SWE-bench Pro tasks specifically Grok 4.5 used about 4.2x fewer output tokens than Opus 4.8 at max effort. Since API bills are token-metered, efficiency multiplies the rate-card advantage.
Pricing
The rate card is $2 per 1M input tokens and $6 per 1M output tokens — confirmed in SpaceXAI's developer docs and matched on OpenRouter. That is well under Claude Opus 4.8 ($5/$25), GPT-5.5 ($5/$30) and Claude Fable 5 ($10/$50), and in the same bracket as mid-tier models like Claude Sonnet 5 ($2/$10 intro). Reasoning effort is adjustable (low/medium/high, default high), which gives another cost lever.
One launch caveat: Grok 4.5 is not yet available in the EU, either in SpaceXAI products or the API console. SpaceXAI says EU availability is expected later in July 2026.
Who is SpaceXAI?
Grok 4.5 is the first model shipped under the SpaceXAI name. SpaceX formally acquired xAI in an all-stock deal that closed on 2 February 2026, valuing the combined company at about $1.25 trillion; SpaceX then went public on 12 June 2026 in a record $75 billion IPO, and the merged AI unit rebranded from xAI to SpaceXAI in early July 2026. The API surface is unchanged — developer docs, console and model IDs still live at x.ai.
Who should use Grok 4.5
- High-volume agentic coding — the token-efficiency numbers were measured on software-engineering benchmarks; cost per resolved task is the metric it wins.
- Routine knowledge work at scale — office/clerical automation, research and drafting where Opus-class quality at a fraction of the price matters more than best-in-class peaks.
- Latency-sensitive products — ~80 TPS puts it at fast-model speeds while staying near flagship quality.
- Not (yet) EU deployments — wait for the EU rollout later in July 2026.
- Not the hardest reasoning tasks — Claude Fable 5 leads every benchmark on SpaceXAI's own chart; if peak capability matters more than cost, the premium models still win.
How Grok 4.5 compares
- Grok 4.5 vs Claude Opus 4.8 — testing the "Opus-class" claim →
- Grok 4.5 vs GPT-5.5 — the value challenger vs OpenAI's flagship →
- Grok 4.5 vs Claude Sonnet 5 — the $2-input bracket showdown →
- Grok 5 — still unreleased; what we know →
Frequently asked questions
When was Grok 4.5 released?
SpaceXAI announced it on 8 July 2026 and opened it to the public on 9 July 2026, after a beta period with early customers. It is the first model released under the SpaceXAI name.
How much does Grok 4.5 cost?
$2 per million input tokens and $6 per million output tokens via the xAI API. SpaceXAI also claims roughly twice the token efficiency of other leading models, compounding the price advantage.
What is Grok 4.5's context window?
500,000 tokens, per SpaceXAI's developer docs — half the 1M window of Claude Opus 4.8 and GPT-5.5.
Is Grok 4.5 really "Opus-class"?
On SpaceXAI's own published chart it beats Claude Opus 4.8 on two of four coding benchmarks and trails on the other two — close to Opus-class on coding, just short of best-in-class, at a much lower price.
Is Grok 4.5 the same as Grok 5?
No. Grok 5 — the reported 6-trillion-parameter frontier model — remains unreleased with no official date. Grok 4.5 is the shipping workhorse.
Is Grok 4.5 available in the EU?
Not at launch. SpaceXAI says EU availability — products and API console — is expected later in July 2026.