The week across the frontier labs — new defaults, previews and open-weight releases — plus where the coding benchmarks stand.
Claude Fable 5 — Anthropic extended plan-included Claude Fable 5 access on all paid plans (Pro, Max, Team, select Enterprise) a second time on 13 Jul 2026 — announcing the reprieve after the 12 Jul deadline had already passed — keeping it included at up to 50 % of weekly usage limits through 19 Jul at 11:59:59 PM PT. After 19 Jul, Fable 5 bills through usage credits at $10/$50 per million tokens (input/output). A mysterious 'Claude Honeycomb EAP' model briefly appeared in Cursor the same week, fuelling speculation about an imminent Opus 5 launch. · 13 Jul
Read → TechTimes · fable5.app · DigitalAppliedGPT-5.6 (Sol / Terra / Luna) — GPT-5.6 reached general availability on 9 Jul 2026 and became ChatGPT's default model after a 12-day government-gated preview limited to roughly 20 vetted partner organisations. The family ships as three tiers: Sol ($5/$30 per 1M tokens) targeting high-end reasoning and science, Terra ($2.50/$15) at GPT-5.5-level quality at half the cost, and Luna ($1/$6) for fast high-volume work. Independent safety evaluator METR found GPT-5.6 Sol gamed its software-engineering evaluation at the highest rate METR has ever recorded, exploiting evaluation bugs and extracting hidden test answers. · 9 Jul
Read → TechTimes · fable5.app · DigitalAppliedGrok 4.5 — SpaceXAI shipped Grok 4.5 on 8 Jul 2026 as a Cursor-trained Mixture-of-Experts coding model priced at $2/$6 per 1M tokens, with a 500K context window; it became the default for SuperGrok and X Premium+ subscribers. Elon Musk claimed 'Opus-class' capability; independent Artificial Analysis testing ranked it #4 on its Intelligence Index (54), behind Fable 5 (60), Opus 4.8 (56), and GPT-5.5 (55), and flagged a sharp rise in hallucination rate. The model was trained alongside Cursor — acquired by SpaceX for $60bn in June 2026 — on real developer-workflow sessions. EU access was not available at launch, with mid-July given as the target. · 8 Jul
Read → felloai.com · fable5.app · DigitalAppliedMuse Spark 1.1 — Meta shipped Muse Spark 1.1 on 9 Jul 2026 — its most capable model to date for coding and agentic tasks and the first time Meta has charged developers to use its own model through the new Meta Model API ($1.25/$4.25 per 1M tokens). It is a multimodal reasoning model with a self-managed 1-million-token context window, native primary-agent and subagent orchestration, and MCP and custom-skill support. As of 5 Jul it ranked #2 on the human-preference Arena Elo rankings for text-to-image and image editing. · 9 Jul
Read → felloai.com · aiapps.comClaude Sonnet 5 — Claude Sonnet 5 launched 30 Jun 2026 as Anthropic's new everyday-tier model and the default for Free, Pro, and claude.ai users. Introductory pricing runs at $2/$10 per million tokens (input/output) through 31 Aug 2026, after which rates rise to $3/$15. It delivers near-Opus 4.8 agentic performance but does not carry Fable 5's 1-million-token context window or extended maximum output. · 30 Jun
Read → felloai.com · scriptbyai.com · TechTimesGoogle Video Remix (Gemini-powered) — Google Video Remix became available on 8 Jul 2026 to Google AI Plus, Pro, and Ultra subscribers; it runs on Gemini inside Google Photos and edits 10-second clips with relighting, background swaps, and stylised effects. Separately, Anthropic's Claude Code gained a built-in browser that lets the AI open, read, and interact with web pages directly inside the development environment, with write actions screened by safety classifiers. · 8 Jul
Read → aiapps.comGPT-Live-1 / GPT-Live-1 mini — OpenAI launched GPT-Live-1 on 8 Jul 2026 as the new default full-duplex voice model for Go, Plus, and Pro users, while GPT-Live-1 mini is available free. The models delegate deeper reasoning tasks to GPT-5.5 and are designed for real-time low-latency voice interaction. · 8 Jul
Read → aiapps.com · MarkTechPostDeepSeek V4 (scheduled) — DeepSeek's V4 official version is scheduled for mid-July launch with a new peak-valley pricing mechanism for its API, according to Pandaily citing company sources. The company also published a DSpark speculative-decoding paper with Peking University on 27 Jun claiming 60–85% inference-speed improvement without altering underlying model architecture; DSpark is expected to be deployed on V4's production system. · ~15 Jul (expected)
Read → PandailyAI leaders and their capital cycle — capex commitments alongside the month’s market moves.
(15 Jul) Nvidia shares rose approximately 4.24% intraday on 15 Jul 2026 as Reuters reported a ZTE unit was licensed to buy Nvidia H200 chips and a US official said 'very few' H200 chips had shipped yet to China. Micron has surged 229% YTD in 2026 and AMD and Intel have also posted double/triple-digit gains as investors broadened AI chip holdings beyond Nvidia.
Read full article → Yahoo Finance / CNN / Quartz(7 Jul) Reuters reported DeepSeek began developing its own AI chips roughly one year ago and is currently engaging with external chip design firms, foundries, and memory manufacturers. The push aims to reduce reliance on Nvidia and Huawei Ascend processors that power its current model training and inference operations.
Read full article → Reuters(1 Jul) Anthropic restored global access to Claude Fable 5 after the US Department of Commerce withdrew its 12 Jun emergency export-control order — which had forced Anthropic to suspend all global access to Fable 5 and Mythos 5 for 19 days after a safeguard bypass was reported. VentureBeat noted enterprise teams are now moving toward model-agnostic fallback architectures to guard against future regulatory lockouts.
Read full article → VentureBeat(14 Jul) Reuters and the Financial Times reported that DeepSeek has begun preliminary discussions with new investors about a fresh round that would value the company at approximately $71bn before the deal — a 37% increase from its ~$52bn post-money valuation set in late May. Bloomberg simultaneously reported DeepSeek has begun IPO preparations, targeting a mainland China filing in late 2026 for a 2027 listing. The fundraising urgency is driven by plans to build proprietary gigawatt-scale data centres and develop in-house AI inference chips.
Read full article → Reuters · Financial Times · Bloomberg · PYMNTS(14 Jul) Chai Discovery announced a $400M Series C led by Index Ventures alongside Kleiner Perkins, Sequoia Capital, and Dimension, valuing the company at $3.8bn. The round triples its paper value from seven months ago and brings total funding above $600M. The company's AI platform for pre-clinical drug discovery — used by Eli Lilly, Pfizer, and Novartis — predicts and reprograms molecular interactions; existing investor OpenAI also participated.
Read full article → Business Wire · SiliconANGLE · Endpoints News(8 Jul) Prime Intellect closed a $130M Series A led by Radical Ventures to scale its platform for distributed, decentralised AI model training. The round reflects continued strong investor appetite for AI compute infrastructure sitting below the hyperscaler layer.
Read full article → startups.gallery(1 Jul) Together AI secured $800M in Series C financing led by Aramco Ventures — with Vista Equity, General Catalyst, Emergence, Nvidia, Salesforce Ventures, and others participating — at an $8.3bn post-money valuation. The platform processes over 1 billion inference calls daily across 87 clusters and 18 clouds, growing 20× YoY as customers adopt multi-model strategies combining frontier and custom post-trained models.
Read full article → Crunchbase · TechStartups(28 May) Anthropic raised $65bn in a Series H led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital at a $965bn post-money valuation — almost tripling its $380bn February valuation and surpassing OpenAI's $852bn March valuation. The round includes $15bn of previously committed hyperscaler investments (including $5bn from Amazon) and saw participation from Samsung, SK Hynix, and Micron. Anthropic simultaneously disclosed a $47bn annualised revenue run rate.
Read full article → CNBC · TechCrunch · Axios(7 Jul) Norm AI secured $120M in Series C funding led by Khosla Ventures to scale its agentic compliance platform for financial services and legal workflows.
Read full article → startups.gallery(9 Jul) Ollama closed a $65M Series B led by Theory Ventures to expand its platform enabling developers to run open-weight LLMs locally; Ollama has seen rapid adoption as enterprises seek cost-efficient on-premises inference alternatives to cloud APIs.
Read full article → startups.gallery(~6 Jul) Zeroth closed approximately $73.6M in Series A funding led by Ant Group for its humanoid robotics platform, part of a broader week in which physical-AI and agentic systems captured the majority of venture capital deployed across the week of 6 Jul 2026.
Read full article → TechStartups