Grok 4.5: xAI's Opus-Class Coder at a Third of the Price
AI News 5 min read

Grok 4.5: xAI's Opus-Class Coder at a Third of the Price

Grok 4.5, released July 8, 2026, is xAI's coding-focused model. It ranks 4th on the Artificial Analysis Intelligence Index (score 54), wins SWE Marathon (29%), and prices at $2/$6 per million tokens with 4.2x better token efficiency than Opus 4.8. Not yet available in the EU.

Sarah Chen
Sarah Chen
Jul 12, 2026

On July 8, 2026, xAI shipped Grok 4.5, the company's self-described "smartest model" and its first built explicitly for coding and agentic work. Elon Musk pitched it in three words: "Opus-class, but faster." The more interesting part is the price tag — Grok 4.5 arrives at roughly one-third the cost of the frontier models it's chasing, and it does the work in a fraction of the tokens.

Here's the honest read: Grok 4.5 is not the single best coding model on the market. But it may be the best value one, and that's a harder thing for competitors to answer.

What xAI actually built

Grok 4.5 was trained alongside Cursor, using real coding-session data, and reinforcement-learned across tens of thousands of NVIDIA GB300 GPUs. xAI leaned hard into "per-token intelligence" — an asynchronous training stack that let agentic rollouts run for hours while learning continued. The payoff shows up less in peak benchmark scores and more in efficiency.

The headline number: on SWE Bench Pro, Grok 4.5 resolves the average task in 15,954 output tokens versus 67,020 for Anthropic's Opus 4.8 — about 4.2× fewer tokens for comparable results. It's served at 80 tokens per second, fast enough that xAI markets it as quicker than most "flash" tier models.

The benchmarks, without the spin

xAI's own launch charts tell a more nuanced story than the marketing. Grok 4.5 leads on some evals and trails on others:

Benchmark Grok 4.5 Best competitor
SWE Marathon (resolution) 29.0% Opus 4.8 — 26.0%
DeepSWE 1.0 (pass@1) 62.0% Fable 5 — 66.1%
Terminal Bench 2.1 83.3% Fable 5 — 84.3%
DeepSWE 1.1 53.0% GPT-5.5 — 67.0%
SWE Bench Pro 64.7% Fable 5 — 80.4%

So: Grok 4.5 wins outright on SWE Marathon and stays within a point on Terminal Bench, but it lands behind Fable 5, GPT-5.5, and Opus 4.8 on DeepSWE 1.1 and SWE Bench Pro. On the aggregate Artificial Analysis Intelligence Index, Grok 4.5 scores 54 and ranks fourth — behind Fable 5, GPT-5.5, and Opus 4.8, but ahead of every Gemini and open-weight model.

That's not a knock. Fourth-best frontier intelligence at a budget price is a genuinely strong position.

The price war is the story

This is where Grok 4.5 gets aggressive. Pricing is $2 per million input tokens and $6 per million output tokens, with cached input at $0.50 per million — a 75% discount. xAI claims that undercuts Opus 4.8 and GPT-5.5 by over 60%.

Now stack the token efficiency on top. If Grok 4.5 needs roughly a quarter of the output tokens to finish a coding task and those tokens cost less, the effective cost gap on real agentic workloads widens well beyond the sticker difference. For teams running coding agents at scale — where output tokens dominate the bill — that compounding is the whole pitch.

The competition isn't "which model tops one benchmark." It's "which model finishes the job for the least money." On that axis, Grok 4.5 is hard to beat right now.

What you can actually do with it

Grok 4.5 is the new default in Grok Build, xAI's agentic CLI, and it's available in Cursor on all plans and through the xAI API console as model ID grok-4.5. xAI is offering free usage for a limited time in Grok Build and Cursor.

Beyond code, xAI is pushing Grok 4.5 into office work — building multi-sheet Excel models with web research baked in, assembling PowerPoint decks using native shapes, and drafting Word documents. Whether those features hold up outside curated demos is the kind of thing worth testing before you trust it with a quarterly report.

A calling curl request looks like this:

curl -s https://api.x.ai/v1/responses \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "grok-4.5", "input": "Find and fix the bug: function median(a){a.sort();return a[a.length/2]}"}'

The catch

Two things temper the enthusiasm. First, Grok 4.5 isn't available in the EU at launch — not in xAI's products and not in the API console — with availability "expected in mid-July." European teams are locked out for now. Second, the benchmark picture means you shouldn't reach for Grok 4.5 expecting it to beat Fable 5 or Opus 4.8 on the hardest software-engineering suites. It won't. It'll get close, cheaper, and faster.

The Bottom Line

Grok 4.5 reframes the coding-model race around cost-per-completed-task instead of leaderboard bragging rights. It ranks fourth on aggregate intelligence, wins a couple of agentic benchmarks outright, and does it all at roughly a third of the price of Opus 4.8 and GPT-5.5 — while burning far fewer tokens. If you're running coding agents at volume and you're outside the EU, it deserves a spot in your evaluation this week. Just don't expect a benchmark sweep; expect a smaller invoice.

More in AI News

GPT-5.6: OpenAI's Sol, Terra, and Luna Go Public
AI News

GPT-5.6: OpenAI's Sol, Terra, and Luna Go Public

OpenAI made its three-tier GPT-5.6 family (Sol, Terra, Luna) generally available on July 9, 2026 after government safety review. Pricing runs from Luna at $1/$6 to Sol at $5/$30 per 1M tokens, with a Sol Fast option at $12.50/$75 on Cerebras. The release adds Programmatic Tool Calling in the Responses API (63.5% fewer tokens, 50.1% fewer turns) and longer prompt caching, but Sol's 64.6% on SWE-Bench Pro still trails Claude Mythos 5 (80.3%).

By Sarah Chen · 5 min · Jul 11, 2026

GPT-Realtime-2.1: OpenAI Adds Reasoning to Its Voice API
AI News

GPT-Realtime-2.1: OpenAI Adds Reasoning to Its Voice API

On July 6, 2026, OpenAI released GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for the Realtime API. The headline change is reasoning in the low-cost mini tier, plus a 25% cut in p95 latency from better caching. The mini holds the prior gpt-realtime-mini price (0 audio in, 0 audio out per 1M) while the full model runs 2/4. Reasoning effort is configurable from minimal to xhigh.

By Sarah Chen · 5 min · Jul 8, 2026

Gemini 3.5 Flash: The Flash Model That Beats Google's Own Pro Tier
AI News

Gemini 3.5 Flash: The Flash Model That Beats Google's Own Pro Tier

Google released Gemini 3.5 Flash on May 19, 2026, at Google I/O. The Flash-tier model beats Gemini 3.1 Pro on coding and agentic benchmarks (76.2% Terminal-Bench 2.1, 83.6% MCP Atlas, 1656 GDPval-AA Elo) while running 4x faster and costing $1.50/$9 per 1M tokens, 40% below 3.1 Pro. It trails Pro on academic reasoning (Humanity's Last Exam, ARC-AGI-2) and dense long-context recall. It powers Gemini Spark, Antigravity 2.0, and is now the default model for the Gemini app and AI Mode in Search.

By Sarah Chen · 5 min · Jul 7, 2026