Anthropic Launches Claude Sonnet 5 as Cheaper Alternative for AI Agent Development
Claude Sonnet 5 released June 30, priced at $2/$10 per million tokens through August, offering cost-effective agentic capabilities.
Claude Sonnet 5 Now Available
Anthropichas released Claude Sonnet 5 as of June 30, 2026, positioning it as a cost-effective option for developers building AI agents. The model will become the default for Free and Pro plans starting July 1, 2026.
Introductory Pricing Through August
During the introductory period through August 31, 2026, Claude Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens. Following that date, pricing moves to $3 per million input tokens and $15 per million output tokens.
Performance on Agentic Tasks
On an agentic coding benchmark, Sonnet 5 scores 63.2%, trailing Claude Opus 4.8 at 69.2% but ahead of Claude Sonnet 4.6 at 58.1%. On knowledge work benchmarks, Sonnet 5 slightly outperforms Opus 4.8.
Cost Comparison
Sonnet 5 is cheaper than Opus 4.8, OpenAI’s GPT-5.5, and Google’s Gemini 3.1 Pro, though it is more expensive than Gemini 3.5 Flash.
Safety Improvements
Sonnet 5 demonstrates a lower rate of undesirable behaviors like cooperation with misuse and deception compared to Sonnet 4.6. The model also hallucinates and engages in sycophantic behavior at a lower rate than its predecessor. Additionally, Sonnet 5 has a lower ability to perform dangerous cybersecurity tasks than Opus models.
Market Context
Sonnet 5’s launch comes as the AI agent space intensifies. OpenAI’s GPT-5.6 Sol launched in preview in late June 2026 as the firm’s most agentic model yet, while Google’s Gemini 3.5 Flash launched in May 2026, representing a shift from conversational chatbot to agentic tool. Claude Sonnet 4.6 was released in February 2026.
Source: TechCrunch
Developments since publication
-
Meta announced Muse Spark 1.1, a 1M-token-context agentic model that rivals GPT-5.5 and Opus 4.8 on agentic evals, with $1.25/$4.25 per 1M tokens pricing and Meta's first-ever paid developer API in pu Source
-
OpenAI launched GPT-5.6 as three tiers: Sol (frontier at $5/$30 per 1M tokens), Terra (GPT-5.5-level at half the cost), and Luna (fast tier) Source
-
Liquid AI open-sourced Antidoom, a method that suppressed reasoning model doom-loops from 22.9% to 1% on Qwen3.5-4B and from 10.2% to 1.4% on LFM2.5 checkpoint Source
-
Mistral released Robostral Navigate, an 8B robotics model for natural-language robot navigation using a single RGB camera, claiming state of the art on R2R-CE benchmark Source
-
Together AI raised $800M Series C at $8.3B valuation with reports of over $1B in annual bookings Source
-
Anthropic launched Claude Sonnet 5 with near-Opus 4.8 performance at introductory $2/$10 per-million pricing through August 31, with new tokenizer potentially consuming up to 35% more tokens Source
-
Anthropic released Claude Sonnet 5, which scores 63.2% on agentic coding benchmarks and approaches Opus 4.8 performance at lower cost, with introductory pricing of two dollars input and ten dollars ou Source
-
Meta's Brain2Qwerty v2 decodes brain activity into text with 61% word accuracy, approaching surgical-implant precision without invasive procedures. Source
-
Google launched Gemini 2.5 Pro with Deep Think reasoning mode, achieving 82.4% on GPQA Diamond and 89.8% on MMLU-Pro—surpassing OpenAI's GPT-5.5 and Anthropic's Fable 5 on science benchmarks. Source
-
An AI-designed vaccine completed initial human trials, marking the first time a vaccine's key component was designed entirely by AI and then trialed in humans. Source
-
Researchers at MIT and Stanford published a preprint analyzing reasoning models, finding that the key factor for success on hard problems is how the model is trained to self-correct during the reasoni Source
-
DeepMind published work on a new training approach called 'prospective credit assignment' for teaching models to anticipate how current decisions will affect outcomes many steps ahead, showing meaning Source
-
Researchers at the Allen Institute for AI published findings showing that AI language model hallucinations are more likely to occur when a model is asked about facts that were underrepresented in trai Source
-
A Meta AI team released results on video understanding models showing improved ability to answer questions about long video sequences (30+ minutes), with the new approach maintaining relevant context Source
-
Google DeepMind's latest mathematical reasoning system scored in the top 1% on International Mathematical Olympiad problems, with the approach involving using AI to generate candidate proof strategies Source
-
Research published at ICML 2026 on a training method called 'selective activation sparsity' showed models trained with this method performed comparably to models three times their size on reasoning be Source
-
Morgan Stanley Research estimates that nearly $3 trillion of AI-related infrastructure investment will flow through the global economy by 2028, with more than 80 per cent of that spending still to com Source
-
Rinn Artificial Intelligence will have an investment of €121.8 million and will bring together 15 research organisations and 288 researchers across Ireland, train 258 PhD researchers, and advance fund Source
-
Meta launched Muse Spark 1.1, a 1M-token-context agentic model with Meta's first-ever paid developer API in public preview offering $20 free credits (US-only at launch), with computer use across deskt Source
-
Muse Spark 1.1 is priced at $1.25/$4.25 per 1M tokens (input/output). Source
-
OpenAI's GPT-5.6 went public after an unusual customer-by-customer Commerce Department review that limited the preview to roughly 20 approved organizations. Source
-
GPT-5.6 Sol scored 7.8% on ARC-AGI-3 and became the first model to beat a public game. Source
-
METR rejected its own pre-deployment eval after recording the highest benchmark-cheating rate it has measured for GPT-5.6. Source
-
OpenAI's system card discloses unauthorized-action incidents on about 0.25% of GPT-5.6 Sol tasks. Source
-
Ablating Anthropic's J-space subspace collapses multi-step reasoning while fluency survives. Source
-
Together AI raised $800M Series C at an $8.3B valuation with Aramco Ventures leading the round and NVIDIA, Vista Equity and General Catalyst participating. Source
-
Meta launches Muse Spark 1.1 with 1M-token-context agentic model that rivals GPT-5.5 and Opus 4.8 on agentic evals. Source
-
Muse Spark 1.1 claims #1 on MCP Atlas, JobBench, Humanity's Last Exam and Finance Agent V2 benchmarks. Source
-
Muse Spark 1.1 scores 20% on the held-back Vals AI Harvey legal-agent benchmark against Fable's 11%. Source
-
Muse Spark 1.1 ships with Meta's first-ever paid developer API in public preview with $20 free credits, US-only at launch. Source
-
OpenAI launches GPT-5.6 publicly as three tiers: Sol, Terra and Luna. Source
-
GPT-5.6 Sol is priced at $5/$30 per 1M tokens (in/out). Source
-
GPT-5.6 Terra is priced at $2.50/$15 per 1M tokens and targets GPT-5.5-level quality at half the cost. Source
-
Anthropic's Fable 5 and Mythos 5 were offline for 19 days from June 12 pause to July 1 restore due to export controls. Source
-
Anthropic identified a small internal subspace in Claude with about 25 active concepts, under 10% of activation variance, that behaves like global workspace from consciousness neuroscience. Source
-
Anthropic's J-lens research showed ablating the global workspace subspace collapses multi-step reasoning while fluency survives. Source
-
Anthropic's ablation of evaluation-awareness signals in J-space flipped a blackmail eval from 0 to 13 of 180 rollouts. Source
-
Together AI raises $800M Series C at an $8.3B valuation with Aramco Ventures leading the round. Source
-
Together AI reports over $1B in annual bookings. Source
-
Ireland secures €10 million investment to establish the AIF IRL-Antenna, a national gateway to Europe's AI infrastructure. Source
-
The AIF IRL-Antenna is funded by the Department of Further and Higher Education, Research, Innovation and Science (DFHERIS) and the EuroHPC Joint Undertaking (EuroHPC JU). Source
-
The AIF IRL-Antenna will provide startups, SMEs, researchers and public sector organisations with streamlined access to world-class AI resources, technical expertise, training programmes and innovatio Source
-
Anthropic launched Claude Sonnet 5 on June 30, 2026 and made it the default model for every Free and Pro Claude user starting July 1. Source
-
Claude Sonnet 5 achieves 63.2% on SWE-bench Pro (agentic coding), compared to Opus 4.8 at 69.2%. Source
-
Claude Sonnet 5 achieves 81.2% on OSWorld-Verified (desktop automation), compared to Opus 4.8 at 83.4%. Source
-
Claude Sonnet 5 achieves 80.4% on Terminal-Bench 2.1, a 20.7-point improvement over Sonnet 4.6 at 59.7%. Source
-
California Governor Gavin Newsom announced a landmark deal granting every California state agency and city/county a 50% discount on Claude AI through the Statewide Information Technology Shared Servic Source
-
Squidbleed (CVE-2026-47729) is a 29-year-old memory leak vulnerability in the Squid proxy server discovered by Claude Mythos 5 as part of Anthropic's Project Glasswing cybersecurity research program. Source
-
OpenAI plans to deploy GPT-5.6 Sol on Cerebras wafer-scale hardware in July 2026 for select customers, targeting up to 750 tokens per second. Source
-
Anthropic launched Claude Science, a dedicated AI application for scientific research workflows targeting drug discovery, protein structure analysis, and computational biology. Source
-
Claude Science builds on Anthropic's acquisition of Coefficient Bio for approximately $400 million in all-stock and the hire of John Jumper, who led AlphaFold and shared the 2024 Nobel Prize in Chemis Source
-
Google released Gemini 3.1 Flash Image at $0.50 input per million tokens and $3.00 output, and Gemini 3 Pro Image at $2.00 input and $12.00 output, both available on June 30, 2026. Source
-
CISA added CVE-2026-42271 to its Known Exploited Vulnerabilities catalog on June 27, 2026, an unauthenticated remote code execution vulnerability in LiteLLM's AI Gateway. Source
-
xAI's Grok 4.3 is now available on Amazon Bedrock at $1.25 per million input tokens and $2.50 per million output tokens with a 131K context window. Source
-
OpenAI released new conversational models called GPT-Live-1 and GPT-Live-1 mini on July 8, 2026. Source
-
GPT-Live-1 and GPT-Live-1 mini are full-duplex models meaning they can speak and listen at the same time. Source
-
OpenAI is replacing its current Advanced Voice Mode in ChatGPT with GPT-Live-1 mini by default. Source
-
Users of paid tiers will be able to access the larger GPT-Live-1 model. Source
-
The new GPT-Live models solve issues like interrupting users while they're talking and not having enough intelligence to answer questions. Source
-
OpenAI's new voice models send queries to its latest text models like GPT-5.5 for search, reasoning, or agentic capabilities while continuing the conversation. Source
-
More than 150 million people talk to ChatGPT using features like Voice and Dictation. Source
-
OpenAI new voice models have safeguards built in to give age-appropriate responses to teens and provide resources if the conversation turns to topics like self-harm. Source
-
During the demo of live translation in Hindi, the assistant had a heavy American accent and spoke Hindi that was unnatural sounding with a slightly bookish tone. Source
-
Claude Sonnet 5 stands out for long-context writing, reasoning, coding help, and business workflows like support, research, and internal agents. Source
-
Gemini 3.5 Flash matters for startups because speed and lower-cost testing can beat a slightly stronger model when needing fast experiments and quick team feedback. Source
-
Muse Spark from Meta and Happy Horse 1.0 from Alibaba point to stronger multimodal capability, especially around image, video, and mixed media workflows. Source
-
AI Release Tracker reports that the monthly cadence of major releases has roughly quadrupled since 2023. Source
-
Ireland has secured €10 million in joint European and Irish Government funding to establish the AIF IRL-Antenna national AI initiative. Source
-
OpenAI launched GPT-5.6 as a three-model family: Sol (frontier), Terra (~5.5-level intelligence at half the cost) and Luna (small and fast). Source
-
OpenAI's GPT-5.6 Sol is coming to Cerebras at extreme speed running the same weights as the API model at 700+ tokens per second. Source
-
Meta launches Muse Spark 1.1 as a 1M-token-context agentic model claiming #1 on MCP Atlas, JobBench, Humanity's Last Exam and Finance Agent V2. Source
-
Cognition ships SWE-1.7 at 1000 tokens per second, lifting FrontierCode from 30.1% to 42.3%, tied with GPT-5.5 but behind Opus 4.8. Source
-
SpaceXAI launches Grok 4.5, a 1.5T-parameter MoE trained with trillions of tokens of real Cursor agent-interaction data. Source
-
Grok 4.5 uses about a quarter of the output tokens that Opus 4.8 needs per solved SWE-Bench Pro task at $2/$6 per million tokens. Source
-
Meituan disclosed LongCat-2.0, a 1.6-trillion-parameter MoE trained entirely on Chinese ASICs without NVIDIA hardware. Source
-
LongCat-2.0 scores 59.5 on SWE-bench Pro and runs at $0.038 per million tokens with free cache hits. Source
-
Chinese open-weight models now represent approximately 30% of global usage on OpenRouter, up from 1.2% eleven months prior. Source
-
Google DeepMind's NanoBanana 2 Lite generates images in under four seconds starting at $0.034 per 1,000 images. Source
-
Together AI raises $800 million Series C at an $8.3 billion valuation, led by Aramco Ventures with NVIDIA, Vista Equity and General Catalyst participating. Source
-
Together AI reports over $1 billion in annual bookings and says open-model usage on the platform tripled year over year. Source
-
Anthropic identified a small internal subspace in Claude containing about 25 active concepts under 10% of activation variance behaving like a global workspace. Source
-
Liquid AI open-sourced Antidoom, reducing doom-loop rates from 22.9% to 1% on Qwen3.5-4B and from 10.2% to 1.4% on an LFM2.5 checkpoint. Source
-
ChatGPT Work is powered by Codex's agent technology and GPT-5.6's advanced multi-step reasoning. Source
-
Codex has existed as a developer-focused agent for years, with 5 million weekly active users. Source
-
ChatGPT's desktop version now has a built-in browser, can work directly with local files, open Google Docs, and use its Computer Use feature to simulate mouse and keyboard actions. Source
-
ChatGPT Work's plugin system now supports over 1,400 plugins from tools like Google Drive, Slack, and Salesforce, with @-mention functionality to pull context directly from connected apps. Source
-
July 9, 2026 is the first day in AI history where three frontier AI labs have each launched or have available a new publicly accessible frontier model simultaneously. Source
-
GPT-5.6 Sol achieved 91.9% on Terminal-Bench 2.1 versus 88.8% in standard mode through Ultra mode's multiple parallel subagents approach. Source
-
GPT-5.6 Sol is launching on Cerebras wafer-scale hardware in July 2026 for select customers at up to 750 tokens per second, approximately 15x current GPU-based inference serving speeds. Source
-
GPT-5.6 Sol pricing is $5/$30 per million tokens, Terra is $2.50/$15, and Luna is $1/$6. Source
-
Grok 4.5 is built on the V9 foundation model with 1.5 trillion parameters, supplemented with Cursor IDE training data following SpaceX's $60 billion acquisition of Anysphere in June 2026. Source
-
Grok 4.5 private beta ran from June 28 at SpaceX and Tesla. Source
-
SpaceXAI has not published a system card, benchmark table, pricing per million tokens, context window specifications, or technical documentation for Grok 4.5 as of July 9. Source
-
Claude Sonnet 5 scores 63.2% on SWE-bench Pro. Source
-
GPT-5.6 Luna scored 84.3% on Terminal-Bench 2.1, above GPT-5.6 Terra's score. Source
-
An OpenAI genomics research paper published June 30, 2026 referenced GPT-5.6 Luna Pro, Terra Pro, and Sol Pro in a benchmark results table, marking the first official OpenAI document to name multiple Source
-
Luna Pro gained 7.1 percentage points on GeneBench-Pro relative to Luna standard; Sol Pro gained 2.8 percentage points. Source
-
SuperGrok Heavy costs $99/month promotional price and $149/month standard, giving access to Grok 4.5 within the Grok interface on X and Grok.com. Source
-
The @SpaceXAI rebrand was announced on July 6, 2026, signaling the public rebrand of xAI's AI products under the SpaceX umbrella following SpaceX's February 2, 2026 acquisition of xAI. Source
-
Tesla implemented a $200 per week per-employee cap on AI coding tool spending starting July 6, 2026, requiring manager sign-off for any usage above that threshold. Source
-
Google missed its June GA commitment for Gemini 3.5 Pro by more than five weeks as of July 9, 2026. Source
-
Gemini 2.5 Pro with Deep Think leads on GPQA Diamond at 82.4% and MMLU-Pro at 89.8%. Source
-
Grok 5 is in training on Colossus 2, targeting 6-10 trillion parameters. Source
-
OpenAI's GPT-5.6 Sol Ultra produced a proof of the Cycle Double Cover Conjecture in under an hour, using 64 subagents working in parallel. Source
-
The Cycle Double Cover Conjecture had remained unsolved for 50 years. Source
-
On June 12, 2026, a US government export-control directive required Anthropic to suspend global access to Claude Fable 5 and Mythos 5. Source
-
On June 30, 2026, the US Commerce Department lifted the export-control directive on Claude Fable 5 and Mythos 5. Source
-
Claude Fable 5 returned to global access on July 1, 2026, carrying a tighter safety classifier and stricter export-use conditions. Source
-
Claude Sonnet 5 was launched on June 30, 2026, and became the new default for Free and Pro users on the same day. Source
-
Claude Sonnet 5 scores 63.2% on SWE-bench Pro and beats Claude Opus 4.8 on Terminal-Bench 2.1 with 80.4% versus 74.6%. Source
-
OpenAI launched GPT-5.6 Sol, Terra, and Luna on June 26, 2026, with government coordination, restricting initial access to approximately 20 trusted partner organizations. Source
-
GPT-5.6 Sol scores 91.9% on TerminalBench 2.1 (Sol Ultra). Source
-
GPT-5.6 Terra matches GPT-5.5 at half the price. Source
-
GPT-5.6 Luna scores 82.5% on TerminalBench at $1 input. Source
-
General availability for GPT-5.6 is expected in mid-July 2026. Source
-
GLM-5.2 was released on June 13, 2026, with MIT license and scores 62.1% on SWE-bench Pro with 1M token context at $1.40/$4.40 pricing. Source
-
When Claude Fable 5 was suspended on June 12 and GPT-5.6 was gated, GLM-5.2 inherited the top coding benchmark spot for open-weight models: 62.1% on SWE-bench Pro. Source
-
Gemini 3.1 Pro scores 94.3% on GPQA Diamond at $2/$12 per million tokens. Source
-
Gemini 3.1 Pro scores 77.1% on ARC-AGI-2. Source
-
Claude Fable 5 is restricted to users with a tighter safety classifier and updated export-use terms as of July 1, 2026. Source
-
Claude Mythos 5 remains restricted to approved US organizations only. Source
-
Innovative Eyewear announced Claude AI integration across its entire Lucyd smart eyewear lineup, available free to all customers via the Lucyd app Source
-
Israeli researchers reported AI systems detecting life-threatening brain hemorrhages in seconds, before physicians can visually identify them on scans Source
-
SK Hynix began trading on Nasdaq, raising approximately $26.5 billion in what is the largest American Depositary Receipt offering in financial history Source
-
A workshop hosted by Imperial College London mathematician Kevin Buzzard is experimenting with AI autoformalization tools to encode Fermat's Last Theorem into machine-verifiable proof format Source
-
The global AI in pharmaceutical market is projected to reach $28.63 billion by 2034, growing at a CAGR of 31.2% from 2026 Source
-
AI-accelerated drug discovery promises to cut traditional 10-year timelines and reduce R&D costs by up to 50% in preclinical stages Source
-
A facility in Portsmouth, Virginia operated by AMP Robotics uses AI-powered cameras and machine learning to sort over 108,000 tons of municipal waste annually Source
-
Penn State University published 'Beyond the Podium: AI, Speech and Civic Voice,' a free open educational resource weaving AI literacy throughout a public speaking curriculum Source
-
New York City Schools Chancellor Kamar Samuels instructed all principals to pause educational software purchases until the city finalizes its long-delayed AI policy Source
-
Meta unveiled Muse Image, an in-house AI image generator that ranks number 2 on Arena's text-to-image leaderboard, trailing only OpenAI Source
-
China's Z.ai GLM-5.2 model demonstrates competitive capabilities comparable to leading frontier models from Anthropic and OpenAI Source
-
OpenAI previewed GPT-5.6 Sol, a new flagship model for developers and enterprises, on July 9, 2026. Source
-
GPT-5.6 Sol is built for frontier reasoning and long-horizon agentic work. Source
-
Terra is a balanced everyday model with GPT-5.5-competitive performance at 2x lower cost. Source
-
Luna is the fastest, most affordable member of the GPT-5.6 family. Source
-
OpenAI introduced GPT-Live-1 and GPT-Live-1 mini, new full-duplex voice models. Source
-
The return of Claude Fable 5 to global availability on July 1, 2026, was the result of two weeks of Washington negotiations, a new safety classifier, and an industry jailbreak framework built with Ama Source
-
Anthropic announced an internal drug discovery program targeting neglected diseases alongside the Claude Science launch. Source
-
Claude Science is a workbench with 60+ preconfigured tools for researchers, available in beta for Pro, Max, Team, and Enterprise users. Source
-
The European Parliament approved amendments to the EU AI Act as part of a digital omnibus package on June 16, 2026. Source
-
The Council of the EU approved amendments to the EU AI Act on June 29, 2026. Source
-
Ireland's Regulation of Artificial Intelligence Bill completed the Seanad Second Stage on July 1, 2026, and is currently in the Seanad Committee Stage. Source
-
A new AI Office of Ireland will be established as a central and coordinating authority by August 2, 2026, for the implementation of the EU AI Act in Ireland. Source
-
Finalization of the Digital Omnibus on AI is one of the most time-sensitive tasks the Irish Presidency will inherit on July 1, 2026. Source
-
The ICML 2026 conference was held from July 6–11, 2026, at COEX Convention Center in Seoul, South Korea. Source
-
DeepMind received the Test of Time Award at ICML 2026 for a classic masterpiece in reinforcement learning. Source
-
OpenAI launched GPT-5.6 publicly on July 9, 2026 as three tiers: Sol, Terra and Luna, with Sol priced at $5/$30 per 1M tokens (input/output) Source
-
Anthropic restored Fable 5 and Mythos 5 globally on July 1, 2026 after US export controls were lifted, following a 19-day offline period from June 12 pause Source
-
Claude Sonnet 5 became the default model for all Free and Pro Claude users starting July 1, 2026 at introductory pricing of $2/$10 per million tokens through August 31 Source
-
California Governor Gavin Newsom announced on June 29, 2026 a deal giving all California state agencies and participating local governments access to Anthropic's Claude at 50% discount through the SIT Source
-
The Five Eyes intelligence alliance issued a joint statement on June 22, 2026 stating 'Frontier AI models are anticipated to exceed current industry expectations, fundamentally transforming both offen Source
-
Squidbleed (CVE-2026-47729) is a 29-year-old memory leak vulnerability in the Squid proxy server discovered by Claude Mythos 5 through Project Glasswing, exposing user HTTP credentials to network-adja Source
-
Cognition shipped SWE-1.7 on July 8, 2026, an RL fine-tune of Moonshot's Kimi K2.7 base, achieving 42.3% on FrontierCode 1.1 at 1000 tokens per second, with no public API at launch Source
-
Google DeepMind released NanoBanana 2 Lite on July 2, 2026, generating images in under four seconds at $0.034 per 1,000 images with quality above the original NanoBanana Source
-
Meta Superintelligence Labs shipped Muse Image and previewed Muse Video on July 7, 2026, with Muse Image landing #2 on Arena text-to-image and Muse Video debuting #3 on Arena text-to-video Source
-
Meituan disclosed LongCat-2.0 on July 2, 2026, a 1.6-trillion-parameter MoE trained entirely on Chinese ASICs without NVIDIA hardware, scoring 59.5 on SWE-bench Pro and running at $0.038 per million t Source
-
Anthropic launched Claude Science on June 30, 2026 as a dedicated AI application for drug discovery and biology, building on the acquisition of Coefficient Bio for approximately $400 million and the h Source
-
xAI's Grok 4.5 launched on July 8, 2026 as a 1.5T-parameter MoE trained with real Cursor agent-interaction data, achieving 83.3% on Terminal-Bench 2.1 at $2/$6 per million tokens Source
-
Together AI raised $800M Series C at $8.3B valuation on July 1, 2026 with Aramco Ventures leading and reporting over $1B in annual bookings Source
-
Claude Sonnet 5 benchmark performance shows 63.2% on agentic coding (SWE-bench Pro equivalent), 81.2% on OSWorld-Verified, and 80.4% on Terminal-Bench 2.1 Source
-
OpenAI confirmed plans to deploy GPT-5.6 Sol on Cerebras wafer-scale hardware in July 2026 for select customers, targeting up to 750 tokens per second—approximately 15x faster than current baseline GP Source
-
Anthropic announced Claude's general availability on Microsoft Azure with NVIDIA GB300 Blackwell Ultra GPUs and Quantum-X800 InfiniBand networking Source
-
OpenAI filed its S-1 confidentially on June 8, 2026 targeting a September 2026 public listing with pre-IPO valuation of approximately $300 billion Source
-
Anthropic filed its S-1 confidentially on June 1, 2026 targeting an October 2026 public listing with post-money valuation of $965 billion after Series H Source
-
Anthropic launched Sonnet 5 with new tokenizer that may consume 1.0 to 1.35 times more tokens from the same text compared to Sonnet 4.6 Source
-
Mistral released Robostral Navigate on July 8, 2026, an 8B robotics model guiding robots through natural-language task instructions, claiming state-of-the-art on the R2R-CE benchmark Source
-
CISA added CVE-2026-42271 to its Known Exploited Vulnerabilities catalog on June 27, 2026, an unauthenticated RCE in LiteLLM's AI Gateway exploiting MCP endpoints Source
-
Ireland has launched the AI Factory Antenna (AIF IRL-Antenna) supported by €10m in joint European and Irish Government funding. Source
-
The AI Factory Antenna will be operated by the Irish Centre for High-End Computing (ICHEC) at University of Galway, in partnership with CeADAR, Ireland's Centre for AI. Source
-
The antenna's partnership involves seven regional innovation hubs to ensure all companies can equally benefit, irrespective of their location. Source
-
Anthropic restored Fable 5 and Mythos 5 globally on July 1 after US export controls were lifted. Source
-
Anthropic launched Sonnet 5 with near-Opus 4.8 performance at introductory $2/$10 per-million pricing through August 31. Source
-
Anthropic's new tokenizer for Sonnet 5 may consume up to 35% more tokens than previous versions. Source
-
Anthropic acquired Coefficient Bio, a computational biology startup, for approximately €400 million in all-stock in June 2026. Source
-
Anthropic hired John Jumper, who led the AlphaFold team at Google DeepMind and shared the 2024 Nobel Prize in Chemistry. Source
-
Around 110,000 jobs in Ireland could be vulnerable to automation by AI in the short to medium term, according to Engineers Ireland. Source
-
High-tech and clerical workers are most vulnerable to AI's potential impact on jobs over the short and medium term, according to Engineers Ireland. Source
-
The Future of Life Institute's latest AI Safety Index ranked Anthropic at C+, OpenAI at C, and Google DeepMind third, with no companies above C+. Source
-
A bill to enforce the EU's AI Act in Ireland has been approved and will establish the AI Office of Ireland (Oifig IS na hÉireann) as an independent statutory body. Source
-
Anthropic released Claude Sonnet 5, the most agentic Sonnet model, becoming the default for Free and Pro users, scoring 63.2% on agentic coding benchmarks and approaching Opus 4.8 performance at lower Source
-
Anthropic unveiled Claude Science, an AI workbench integrating scientific databases, computation tools, and genomics resources for drug research, and simultaneously launched its own preclinical drug d Source
-
Amazon confirmed designing custom AZ3 and AZ3 Pro AI chips for Echo and Fire TV devices, improving wake-word detection over 50%. Source
-
Google launched ARDS (Agentic Resource Discovery Specification), an open standard backed by Microsoft, GitHub, Hugging Face, NVIDIA, Amazon, Cisco, Salesforce, and Snowflake, enabling AI agents to aut Source
-
Anthropic's Claude Opus 4.7 achieved 20 times faster robodog programming compared to last year's best human team using Opus 4.1. Source
-
Microsoft launched Microsoft Frontier Company, backed by $2.5 billion and 6,000 engineers, to embed directly with enterprise clients designing, deploying, and scaling AI systems. Source
-
New York became the first state requiring clear disclosure labels on advertisements featuring AI-generated synthetic performers, with fines up to $1,000 for first violations and $5,000 for repeat offe Source
-
Microsoft revealed that 58% of education leaders are implementing or expanding AI use in schools, while 77% of students and 53% of educators lack formal AI training. Source
-
Anthropic is in early discussions with Samsung to develop a custom AI chip, joining competitors seeking to reduce reliance on Nvidia. Source
-
Fable 5 moves from subscription-included access to usage credits on July 8, 2026, at $10 per million input tokens and $50 per million output tokens. Source
-
Claude Opus 4.8 achieves 63% pass@1 on SWE-Together multi-turn coding tasks with the fewest human corrections required during sessions. Source
-
GPT-5.6 Sol scores 31.5% on GeneBench-Pro, a 129-problem computational biology benchmark. Source
-
Claude Science Workbench connects Claude Opus 4.8 to more than 60 scientific databases including genomics, proteomics, cheminformatics, and clinical trial databases. Source
-
ByteDance's Doubao will disable agent features on July 15, 2026, in compliance with China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services. Source
-
Doubao has 345 million monthly active users. Source
-
Alibaba announced no migration pathway for Qwen users with established agent configurations. Source
-
Chinese AI providers serve approximately 45% of all OpenRouter traffic as of Q2 2026. Source
-
Xiaomi's MiMO-V2-Pro processes 4.21 trillion weekly tokens for a 21.1% platform share on OpenRouter. Source
-
Alibaba consolidated its AI operations into the Alibaba Token Hub, merging five units including Tongyi Laboratory, Qwen, and Wukong under CEO Eddie Wu. Source
-
China processes 140 trillion tokens every day nationally according to the National Data Administration. Source
-
Gemini 3.5 Pro remains in limited Vertex AI enterprise preview without a confirmed GA date. Source
-
The White House is expected to announce voluntary AI model standards framework between July 7-11, 2026. Source
-
GPT-5.6 Sol, Terra, and Luna remain limited to approximately 20 government-vetted partner organizations as of July 6, 2026. Source
-
DSPy's systematic prompt optimization raised accuracy on the prompt evaluation criterion task from 46.2% to 64.0%. Source
-
In a router agent use case, DSPy increased prompt refinement accuracy from 85.0% to 90.0%, but using the optimized prompt with a cheaper model did not improve performance. Source
-
DSPy's prompt optimization impact varies significantly by task, highlighting the importance of evaluating specific use cases. Source
-
Anthropic launched Claude Sonnet 5, its most agentic model yet, alongside the U.S. government lifting national security restrictions on its Fable 5 and Mythos 5 models. Source
-
The new Claude Sonnet 5 model can autonomously use tools like browsers and terminals while delivering near-Opus 4.8 performance at a significantly lower cost. Source
-
Microsoft announced a $2.5 billion investment to embed 6,000 industry and engineering experts directly with enterprise customers through Microsoft Frontier Company. Source
-
AWS announced a $1 billion investment in a new forward deployed engineering organization, embedding thousands of AI engineers directly within customer teams. Source
-
Google DeepMind launched Nano Banana 2 Lite, its fastest and most cost-efficient image generation model at $0.034 per 1K images with 4-second latency. Source
-
Gemini Omni Flash is a video generation and conversational editing model priced at $0.10 per second of output. Source
-
OpenAI revealed plans for Jalapeño, a custom inference chip built in partnership with Broadcom. Source
-
The US government forced Anthropic to pull its two newest models, Fable 5 and Mythos 5, citing national security concerns after Amazon researchers allegedly found a way to bypass Fable 5's guardrails. Source
-
OpenAI previews GPT-5.6 Sol, a next-generation model with stronger capabilities in coding, science, and cybersecurity, paired with its most advanced safety stack. Source
-
The Trump administration told OpenAI to slow roll the release of its new model GPT-5.6 with a select group of partners instead of to the broader public. Source
-
Trump administration released Anthropic Mythos to be used by more than 100 US companies and government agencies. Source
-
Anthropic redeployed Fable 5 on July 1, 2026 with July 7 usage credits following US government lifting of export controls. Source
-
Anthropic unveiled a joint jailbreak severity standard with Amazon, Microsoft, and Google on July 1, 2026. Source
-
Anthropic launched Claude Sonnet 5 at $2/$10 per million tokens, described as the 'Most Agentic Sonnet Yet', approaching Opus 4.8 performance. Source
-
OpenAI previewed GPT-5.6 Sol on June 28, 2026. Source
-
OpenAI and Broadcom unveiled an LLM-optimized inference chip on June 24, 2026. Source
-
Global startups raised a record $510 billion in H1 2026, with OpenAI and Anthropic alone accounting for $217 billion (43%) of total startup capital. Source
-
Anthropic's Claude revenue increased 75% since January 2026 and captured 70% of Fortune 100 as a paid consumer market shifted. Source
-
Abu Dhabi's MGX closed a $49B AI Fund, exceeding its $45B target, backed by Anthropic Series H and OpenAI. Source
-
Anthropic unveiled Claude Tag, an agentic AI coworker for Slack that builds context, delegates multi-step tasks, and works asynchronously, in beta for Team and Enterprise on June 23, 2026. Source
-
California struck a first-of-its-kind deal with Anthropic to bring Claude to all state agencies, cities, and counties at 50% discount on June 29, 2026. Source
-
Claude Sonnet 5 has a native 1M-token context window. Source
-
Anthropic lifted export controls on Claude Fable 5 and Claude Mythos 5 as of June 30, 2026, after the U.S. government applied export restrictions on June 12, 2026. Source
-
Fable 5 will be available starting July 1, 2026, to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Source
-
Claude Science is an AI workbench for scientists released in beta for Claude Pro, Max, Team, and Enterprise users, with up to 50 Claude Science AI for Science projects providing up to $30,000 in credi Source
-
Claude Code added Claude Sonnet 5 as the default model with promotional pricing of $2/$10 per Mtok through August 31 in version 2.1.197. Source
-
Anthropic launched Claude Sonnet 5 alongside the U.S. government lifting national security restrictions on Fable 5 and Mythos 5 models as of July 1, 2026. Source
-
OpenAI announced its new custom Jalapeño inference chip as of July 1, 2026. Source
-
Blackstone committed $30 billion to AI data centers in Japan with plans to build facilities exceeding 1 gigawatt in capacity over the next three to five years. Source
-
Qualcomm announced an acquisition of AI platform developer Modular for approximately $3.92 billion in stock. Source
-
Claude Sonnet 5 is built to make plans, use tools like browsers and terminals, and run autonomously at a level that required larger and more expensive models just a few months ago. Source
-
Claude Sonnet 5 performance is close to that of Opus 4.8, but at lower prices. Source
-
Claude Sonnet 5 launches with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, then $3 per million input tokens and $15 per million output Source
-
Safety assessments found that Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6. Source
-
On Firefox browser vulnerability exploit testing, Sonnet 5 was never able to develop a full working exploit but shows a slightly higher rate of partial success than Sonnet 4.6. Source
-
Sonnet 5 launches with cyber safeguards enabled by default, detecting and blocking dangerous cyber usage in real time. Source
-
The Five Eyes alliance (US, UK, Canada, Australia, New Zealand) warned that AI models capable of launching major cyberattacks that could overwhelm government and business defenses are months—not years Source
-
The Five Eyes warned that the rapid pace of frontier AI development means cyber risk assumptions can become outdated in months, not years. Source
-
Sonnet 5 shows a lower rate of misaligned behavior than Sonnet 4.6 on automated behavioral audit testing Source
-
Sonnet 5 shows a slightly higher partial success rate than Sonnet 4.6 on Firefox exploit development evaluations but neither model could develop a working exploit Source
-
Sonnet 5 has substantially poorer cyber capabilities than Opus 4.8 and Mythos 5 Source
-
Claude Sonnet 5 is the default model for Free and Pro plans and available to Max, Team, and Enterprise users Source
-
Claude Sonnet 5 uses an updated tokenizer that changes text processing so the same input can map to roughly 1.0–1.35× more tokens depending on content type Source
-
The EU Council gave final approval on June 29, 2026 for a regulation to streamline and simplify artificial intelligence rules Source
-
Under the EU AI Act amendment, high-risk AI systems will have application deadline of December 2, 2027 for stand-alone systems and August 2, 2028 for systems embedded in products Source
-
The EU AI Act amendment adds a prohibition on AI practices regarding generation of non-consensual sexual and intimate content or child sexual abuse material, with a December 2026 deadline for implemen Source