YFarmXIntelligence for an accelerating world

AI Hub

Explore the future
of artificial intelligence

Models, agents, infrastructure and policy: what actually changed, without the hype.

62
AI models & tools explained
60
AI stories published
31
verified AI events ahead
NVDANVIDIA$200.75up2.93%
MSFTMicrosoft$464.72up3.02%
GOOGLAlphabet$356.13up6.73%
AMDAMD$476.15down1.90%

Nasdaq · 02 Aug 2026, 07:56 UTC · refreshed every six hours

Model of the weekKimi K3: the open-weight model that puts China at the frontier

Latest AI news

All news →
47 more AI stories in the Newswire →

State of play

TIME-SENSITIVElast updated 31 July 2026

LLMsClaude Opus 5 landed on 24 July at half Fable 5's price and near its capability, and Kimi K3's open weights now sit just behind the same flagship.$5/$25Opus 5 per M tokens50%of Fable 5's price1.4TBK3 open weights
  1. 30 JulOpenAI cuts GPT-5.6 Luna by 80% to $0.20 and $1.20 per million tokens and Terra by 20% to $2 and $12. Sol stays at $5 and $30.
  2. 27 JulKimi K3's full weights go public, roughly 1.4TB even at four-bit precision. It ranks fourth of 189 models on the Artificial Analysis index, behind only Fable 5 and GPT-5.6 Sol, and first on Arena's Frontend Code board ahead of both.
  3. 24 JulClaude Opus 5 lands at $5 and $25 per million, the same as the Opus 4.8 it retires and half of Fable 5, topping most rows of Anthropic's own table. Fable 5 stays the ceiling: Anthropic reserves it for work that needs the highest capability available, and it keeps narrow leads on Humanity's Last Exam without tools and on FrontierCode.
  4. 21 JulGemini 3.6 Flash ships while Gemini 3.5 Pro stays in partner testing.
  5. 16 JulMoonshot's Kimi K3 arrives at 2.8 trillion parameters.
  6. 9 JulGPT-5.6 goes general in three tiers, Sol, Terra and Luna, with Meta's Muse Spark 1.1 the same day.
  7. StandingSonnet 5 remains the everyday default at $3 and $15 with a 1M-token window.
Explore
Image GenerationGPT Image 2 still leads the arena by a record margin, and it now plans or searches the web before it draws.#1GPT Image 2, arena$0.034per image, NB2 Lite4Knative, NB Pro
  1. 31 JulGPT Image 2 still leads the arena by a record margin, and plans or searches the web before it draws.
  2. 30 JunNano Banana 2 Lite becomes the fast tier at roughly $0.034 an image; Nano Banana Pro keeps precise editing and native 4K.
  3. 24 JulMidjourney V8.2 ships and takes the default seat from V8.1; Personalization profiles are the headline upgrade.
  4. StandingFLUX.2 is the strongest open-weight family; Firefly Image 5 wins wherever licensing decides the tool.
Explore
Video GenerationSeedance 2.5 will generate 30 seconds in a single pass, and Sora 2's API switches off on 24 September.30ssingle pass, Seedance 2.524 SepSora 2 API off4K/60Kling 3.0
  1. 16 JulByteDance opens Seedance 2.5's API with native 30-second single-pass clips, the longest one-shot generation anyone offers.
  2. 24 SepSora 2's API switches off.
  3. StandingGemini Omni Flash tops text-to-video while Seedance 2.0 keeps the image-to-video and editing crowns.
  4. StandingKling 3.0 is the value pick at native 4K and 60fps; Veo 3.1 the safest Western choice, with native dialogue and SynthID provenance.
Explore
Agentic Frameworks / RoutingHermes Agent's Quicksilver release cut time to first token by roughly 80%, and Nous Research, which builds it, is in talks at $1.5B.80%faster first token245kservers exposed$1.5BNous talks
  1. 20 JulHermes Agent's Quicksilver release cuts first-turn time to first token by roughly 80%. Finishing a task still writes the agent a reusable skill, so it improves with use.
  2. 13 JulNous Research, which builds Hermes, is in talks at a $1.5B valuation.
  3. 9 JulOpenAI's ChatGPT Work joins, shipping finished documents, spreadsheets and web apps.
  4. JulyClaw Chain chains four OpenClaw flaws across roughly 245,000 reachable servers. All four are patched.
  5. StandingOpenClaw is the fastest-growing open-source project in GitHub history, Cowork leads desktop agents, and MCP is the universal tool standard.
Explore
AI Coding & Dev ToolsOpus 5 became Claude Code's default on 24 July at the price of the model it replaced, with Fable 5 still the ceiling above it.$2.5BClaude Code run-rate9 mthsto get there18%adoption, tied
  1. 24 JulOpus 5 becomes Claude Code's default at the same price as the model it replaced. Fable 5 is still what you reach for when a task needs the highest capability Anthropic sells.
  2. StandingClaude Code is the fastest-growing developer product on record, at a reported $2.5B revenue run-rate inside nine months.
  3. StandingCursor and Claude Code are tied near 18% adoption each; Copilot's larger installed base has stopped growing.
  4. JuneWindsurf re-emerges as Devin Desktop under Cognition, and Codex runs on the GPT-5.6 Sol family.
Explore
AI CompaniesAnthropic priced Opus 5 at half its own Fable 5 on 24 July, eight weeks after closing $65B at a $965B valuation.$965BAnthropic valuation$65BMay round$47Breported run-rate
  1. 24 JulAnthropic releases Claude Opus 5 at half what it charges for its own Fable 5, undercutting the flagship on price while leaving it at the top of the range.
  2. 15 JulChina's anthropomorphic-agent rules take effect, taking Doubao's and Qwen's human-like companions offline.
  3. 9 JulOpenAI takes GPT-5.6 general and launches ChatGPT Work beside it.
  4. 6-8 JulxAI becomes SpaceXAI after February's merger into SpaceX, and ships Grok 4.5.
  5. 28 MayAnthropic closes a $65B round at a $965B valuation, on a run-rate near $47B.
Explore
Learning & ResourcesPublished benchmark scores saturate within months of release, so the live arenas and usage boards are what practitioners now read.
  1. StandingBenchmarks saturate within months of release, so a published score dates fast. The arenas and usage boards track what people actually reach for.
  2. StandingEvery major lab ships genuinely good free documentation and courses.
  3. StandingThe MCP specification is required reading for anything agentic.
Explore