SpaceXAI released Grok 4.5 to the public today, July 9, following a limited private beta with SpaceX and Tesla engineering teams. Built on a new 1.5-trillion-parameter "V9" foundation and fine-tuned on real developer session data from Cursor — the coding editor SpaceX is acquiring for roughly $60B — the model is priced at $2 input / $6 output per million tokens, well below Claude Opus 4.8's $5/$25 and GPT-5.6 Sol's $5/$30.
Elon Musk called it "an Opus-class model, but faster, more token-efficient and lower cost," though his more precise internal comparison put it closer to Opus 4.7. Of the four benchmarks SpaceXAI chose to publish, Grok 4.5 beats Opus 4.8 on two (DeepSWE 1.0, Terminal-Bench 2.1) and loses on two (DeepSWE 1.1, SWE-Bench Pro) — a genuine split, not the "beats Opus" framing that spread across launch-day coverage. Independent evaluator Artificial Analysis ranks it #4 overall on its Intelligence Index, behind Claude Fable 5, Opus 4.8 and GPT-5.5.
The clearer edge is efficiency: xAI reports Grok 4.5 resolves SWE-Bench Pro tasks using roughly 4.2x fewer output tokens than Opus 4.8 in "max" mode, at about 80 tokens per second. None of the launch-day benchmark numbers have independent third-party verification yet.
OpenAI's three-tier GPT-5.6 family — Sol ($5/$30 per million tokens), Terra ($2.50/$15) and Luna ($1/$6) — moved from a roughly 20-organization vetted preview to full public availability today, July 9, after the Commerce Department's Center for AI Standards and Innovation completed additional testing requested under a June 2 executive order on AI safety review.
This is the same day Grok 4.5 launched publicly — the first time since Claude Fable 5's export-control suspension began June 12 that every major U.S. frontier lab has a model publicly available simultaneously. OpenAI says the gpt-5.5-latest endpoint will not auto-migrate; developers need to pin explicit model IDs (gpt-5.6-sol, -terra, -luna) to pick up the new prompt-caching system.
Terra, priced to match GPT-5.5-class performance at half the cost, is expected to see the fastest early enterprise adoption given its price parity with several near-frontier competitors now live the same week.
SambaNova Systems has closed a $1B Series F led by General Atlantic, with participation from T. Rowe Price, Capital Group, BlackRock-managed funds, Intel Capital and the Qatar Investment Authority, pushing its valuation to $11B. The round comes months after reports the Palo Alto chipmaker was exploring a sale amid fundraising struggles.
Alongside the raise, JPMorgan Chase confirmed it has selected SambaNova as an inference infrastructure partner, deploying its SN40 and SN50 chip systems to support the bank's generative AI workloads — a marquee enterprise win as banks scale up AI inference spend.
The new capital will go toward compute capacity, global deployment and continued development across SambaNova's chip, systems and software stack, positioning it as a smaller but resurgent Nvidia inference competitor.
Google has pushed Gemini 3.5 Pro past its original I/O commitment window, citing token-efficiency issues and coding-performance gaps flagged by enterprise testers during extended agentic tasks. As of today, the model remains in limited Vertex AI enterprise preview with no confirmed general-availability date.
The delay is increasingly conspicuous: today marks the first day every other major U.S. frontier lab — OpenAI, Anthropic and now SpaceXAI — has a current flagship model publicly accessible at the same time. Gemini 2.5 Pro with Deep Think remains available separately through the Gemini API and AI Studio.
Each additional day of delay narrows Google's window to respond competitively on pricing and benchmark positioning before rivals lock in developer mindshare for the current release cycle.
Prime Intellect closed a $130M Series A on July 8, led by Radical Ventures with Nvidia Ventures and Intel Capital participating, pushing its valuation to $1B, TechCrunch reported. The company sells computing power and training tooling that lets other companies train their own AI agents without depending on OpenAI, Anthropic or Google's frontier stacks.
The round lands in the same 24-hour window as several other AI infrastructure and agent-training raises, part of a broader pattern of capital moving toward the "picks and shovels" layer beneath the frontier-model race rather than new foundation models themselves.
Prime Intellect's pitch — democratized training infrastructure for teams that want agent capability without frontier-lab dependency — positions it as a bet on decentralized compute at a moment when Nvidia GPU access remains the industry's tightest bottleneck.