Kimi K3 rebuilt Google Maps 3D in 90 minutes, two prompts
Moonshot's 2.8T open-weight model cloned Google Maps 3D — extrusions, sun shadows, flyTo — from two prompts.
LLMs
Model releases, benchmarks and capabilities. 106 stories, page 3 of 5.
Opus 5 lands: Fable 5 smarts at half the price
Anthropic's Opus 5 targets GPT-5.6 Sol on coding and agents while undercutting it on output cost.
Anthropic's Opus 5 undercuts Fable 5 at half the price
Opus 5 matches frontier rivals at ~$5/$25 per million tokens — half what Fable 5 charges.
Opus 5 beats Fable 5 at physics sims — for half the cash
Anthropic's Opus 5 matches Fable 5's output quality on 3D destruction tests while costing 2x less per run.
Kimi K3 is the first open model to crack 2.8T params
Moonshot AI's Kimi K3 becomes the largest open-weight model ever, tripling its predecessor and topping Claude Opus 4.8 on coding.
Kimi K3 crashes the Fable 5 vs GPT-5.6 party at 1/4 the price
Moonshot's 2.8T open-weight K3 lands 1 in code, near-ties the frontier, and undercuts Fable 5 by 3x.
Alibaba's 2.4T Qwen3.8-Max lands second only to Fable 5
China's two biggest labs shipped frontier-scale models in one week — and both immediately hit the compute wall.
Alibaba's 2.4T Qwen3.8 Claims No.2 Spot, Behind Only Fable 5
Alibaba previewed a 2.4T-parameter multimodal Qwen3.8-Max and says only Fable 5 beats it — with zero benchmarks published.
Kimi K3 Is So Popular Moonshot Had to Stop Selling It
Moonshot paused new Kimi K3 subscriptions after 48 hours of demand pushed its GPU clusters to the edge.
Anthropic extends free Fable 5 access until July 19
Claude Fable 5 stays available inside Claude subscriptions for another testing window, now through July 19.
Grok 4.5 tops pro-work benchmarks at a bargain $2/M
SpaceXAI's Grok 4.5 beats GPT-5.5 and Claude Opus 4.8 on real professional tasks — for just $2 per million input tokens.
Grok 4.5 hits free X accounts — grab it before it's gated
SpaceXAI's new Grok 4.5 is free right now via Grok Build, no Premium+ needed. The window won't stay open.
OpenAI ships GPT-5.6: meet Sol, Terra & Luna
New naming splits generation from tier. Terra is 2x cheaper than GPT-5.5. Public launch gated by a US gov review.
OpenAI's GPT-Live can now listen and talk at the same time
New full-duplex voice models replace ChatGPT Voice — GPT-Live can interrupt, pause, and think in real time.
Grok 4.5 lands at $2/$6 — and undercuts Opus 4.8 ~17x
xAI's new Grok 4.5 is Opus-class but far cheaper: 4x lower per-token cost and 4.2x fewer tokens per task.
Grok 4.5 ships: 500K context, $2/$6 per million tokens
xAI's Grok 4.5 hits API, Console and Grok Build with a 500K window and aggressive frontier-model pricing.
OpenAI's GPT-Live lets ChatGPT talk and listen at once
GPT-Live and GPT-Live mini bring full-duplex voice to ChatGPT, delegating hard questions to GPT-5.5.
Musk: 2T-parameter Grok finishes training this month, ships in August
xAI's next Grok jumps to 2 trillion parameters — training wraps in July, public launch targeted for August.
Anthropic extends free Fable 5 access through July 12
Claude's Fable 5 stays in Pro/Max subscriptions a few days longer before shifting to usage credits.
Xiaomi's MiMo-V2-Pro hits OpenRouter — Opus benchmarks, budget price
Xiaomi's 1T+ param frontier model lands on OpenRouter at $1/$3 per million tokens, with coding that beats Claude Sonnet.
Google's Deep Research lands in Gemini API with MCP and native charts
Google upgraded Deep Research in the Gemini API with MCP support, native chart generation, and a Max tier for heavy async tasks.
Google's Deep Research hits the Gemini API with MCP & native charts
Deep Research and its Max variant are now API-accessible with MCP support, proprietary data grounding, and built-in chart generation.
Google Quietly Updated Deep Research — Max Version Incoming
New model IDs spotted: deep-research-max-preview-04-2026 scores 93.3% on DeepSearchQA and 54.6% on HLE.
Best local AI stack for your $1K GPU — 2026 shortlist
24GB VRAM (4090, 3090, or Mac) now runs frontier-class AI locally. Here's the model lineup that actually delivers.