The week across the frontier labs — new defaults, previews and open-weight releases — plus where the coding benchmarks stand.
GPT-5.6 Sol / Terra / Luna — OpenAI rolled out GPT-5.6 globally on 10 Jul after a two-week US government safety review. The three-tier family — Sol ($5/$30 per 1M tokens), Terra ($2.50/$15), Luna ($1/$6) — introduces 'max' and 'ultra' reasoning modes. Sol sets new state-of-the-art on Agents' Last Exam (53.6%), BrowseComp (92.2%), and the AA Coding Agent Index (80), while Terra matches GPT-5.5 at 2× lower cost. GPT-5.4 retires 23 Jul. · 10 Jul
Read → OpenAI · iClarified · SpaceDailyGrok 4.5 — SpaceXAI launched Grok 4.5 on 8 Jul — its first model built on the 1.5T-parameter V9 MoE foundation and trained alongside Cursor developer session data. Priced at $2/$6 per 1M tokens (60 %+ cheaper than Opus 4.8), it achieves 64.7 % on SWE-Bench Pro and 83.3 % on Terminal-Bench 2.1, and resolves tasks using 4.2× fewer output tokens than Opus 4.8. EU availability expected mid-July. · 8 Jul
Read → SpaceXAI · Axios · TechCrunchMuse Spark 1.1 — Meta released Muse Spark 1.1 on 9 Jul alongside the public preview of the Meta Model API — the first time Meta has charged for model inference ($1.25/$4.25 per 1M tokens, $20 free credits). Built by Meta Superintelligence Labs, the multimodal reasoning model has a 1M-token context window with native context compression, beats Opus 4.8 on agentic benchmarks (JobBench 54.7 vs 48.4; MCP Atlas 88.1), and supports text, image, video, PDF, and audio input. Zuckerberg posted on X for the first time in three years to announce it. · 9 Jul
Read → Meta AI Blog · TechCrunch · ReutersChatGPT Work — Co-launched with GPT-5.6 on 9-10 Jul, ChatGPT Work merges the Codex coding tool into the main ChatGPT desktop app (macOS and Windows), enabling long-horizon agentic tasks. The 'ultra' setting coordinates four parallel agents. GPT-5.4 remains the fast default in standard chat; Work/Codex surfaces Terra and Sol per plan tier. · 10 Jul
Read → OpenAI · SpaceDailyAI leaders and their capital cycle — capex commitments alongside the month’s market moves.
(8 Jul) Since peaking on 14 May, Nvidia shares have shed ~16%, wiping ~$1T in market cap; at 18× forward earnings the stock now sits below the broader S&P 500 multiple (~20×). Analysts note rotation into memory (Micron +229% YTD) and storage names rather than deteriorating fundamentals — Nvidia holds ~97% server GPU share.
Read full article → Yahoo Finance / Quartz(May–Jun 2026) Amazon, Alphabet, Microsoft, and Meta plan $635–665bn combined (some analysts cite $725bn inclusive of updated Q1 guidance uplifts), with ~75% directly tied to AI GPUs, data centers, and custom chips. Goldman Sachs raised its 2025–2030 combined estimate to $5.3T.
Read full article → Yahoo Finance · Tom's Hardware · TechCrunch(10 Jul) As part of the GPT-5.6 launch, OpenAI is deploying Sol on Cerebras hardware targeting 750 tokens per second — aimed at latency-sensitive enterprise workflows. Access is initially limited to select customers as capacity scales.
Read full article → OpenAI · VentureBeat(31 Mar 2026) OpenAI closed $122bn in committed capital at an $852bn post-money valuation — up 70% from its $500bn October 2025 tender. Amazon anchored with $50bn (of which $35bn is contingent on IPO or AGI), Nvidia and SoftBank each contributed $30bn, co-led by SoftBank alongside a16z, D.E. Shaw, MGX, TPG, T. Rowe Price, and others; $3bn raised via retail bank channels. OpenAI reports $2bn/month revenue, 900M weekly active users, and is preparing for a potential 2026 IPO.
Read full article → Bloomberg · CNBC · OpenAI · TechCrunch · Forbes(1 Jul 2026) Together AI closed $800M in Series C funding led by Aramco Ventures, with Nvidia, Vista Equity Partners, and General Catalyst also participating, at a post-money valuation of ~$8.3bn. Annual bookings crossed $1.15bn in Q2 2026. The company plans to grow public cloud capacity 50× over five years, positioning itself as the leading open-source inference platform.
Read full article → Crunchbase · StartupHub.ai · TechStartups(2 Jul 2026) Crusoe, the AI-focused cloud and data center operator, was reported in talks to raise $3bn — a round that would triple its valuation to roughly $18bn. The company's compute capacity targets AI training and inference workloads, with a differentiated stranded-gas power strategy.
Read full article → StartupHub.ai(~1 Jul 2026) Kling AI, the Chinese video-generation AI company, closed a $2bn funding round backed by General Atlantic at an $18bn valuation. The round underscores surging investor demand for specialized generative video infrastructure outside the US.
Read full article → StartupHub.ai(6 Jul 2026) Zeroth raised approximately $73.6M in Series A funding for humanoid robotics, led by Ant Group. The round is part of a broader pattern of Chinese strategic capital backing physical-AI and humanoid companies, alongside Yingzhi XBOT's $56M Series B the same week.
Read full article → TechStartups(Q1 2026 earnings) Amazon guides $200bn, Alphabet $180–190bn, Microsoft ~$190bn annualized, Meta $125–145bn for calendar 2026 AI infrastructure. Goldman Sachs raised its combined 2025–2030 capex estimate for the four to $5.3T. Free cash flow is turning negative for several companies: Oracle already negative; Amazon crossing; Alphabet expected to hit zero around Q1 2027.
Read full article → Yahoo Finance · CNBC · Tom's Hardware · TechCrunch(ongoing 2026) The OpenAI-SoftBank-Oracle Stargate project targets $500bn in AI infrastructure investment through 2029. Eight data centers in Abilene, TX are under construction with the final building on track for end-2026. The project remains the largest single AI infrastructure commitment on record.
Read full article → TechCrunch