AI News — July 18, 2026

Model releases

• [2026-07-16] Moonshot AI — Kimi K3 launches as the largest open-weight model ever announced and takes #1 on the external Frontend Code Arena, ahead of Fable 5 and GPT-5.6 Sol. Moonshot shipped Kimi K3 via its API (api.moonshot.ai, model id kimi-k3): a sparse Mixture-of-Experts model with 2.8 trillion total parameters but only ~50B active per token (896 experts, 16 routed), a 1M-token context window, and multimodal input (text, images, video). On Arena.ai’s Frontend Code Arena — an external, human-preference leaderboard — K3 took #1 with 1,679 points, ahead of Fable 5 (1,631) and GPT-5.6 Sol (1,618), winning six of the board’s seven domains and losing only Gaming to Fable 5. Moonshot’s own benchmark table is more modest: it reports K3 topping Opus 4.8 and GPT-5.5 while still trailing Fable 5 and GPT-5.6 Sol on most suites, so the Frontend Code Arena result is a genuine external SOTA on that specific board, not an across-the-board lead. API pricing is $3 / $15 per million input/output tokens (cache-hit input $0.30) — roughly Sonnet-tier pricing for a model Moonshot frames as Opus-4.8-class. It matters because a Chinese lab now holds the top spot on a widely watched frontend-coding leaderboard with a model whose weights are slated to go public, landing on the exact coding axis where the US flagships compete — and it does so in the same window Google’s Gemini 3.5 Pro slipped for coding shortfalls. The catch: the open weights aren’t out yet — Moonshot says they release July 27, so today K3 is a proprietary API with an open-weights promise attached, and independent evals beyond the frontend arena are still thin. (source, analysis)

  Because K3 is reachable through a standard REST API today — with cache-hit input as low as $0.30 per million — a team can drop it into an existing coding harness this week and benchmark it on their own frontend tickets rather than trusting a single external leaderboard or waiting for the July 27 weights.

  For Software Developers: Since K3 accepts image input, a developer rebuilding a UI from a Figma export can pass the screenshot straight into the prompt and use the 1M-token window to hold the whole design-system token file and existing component code alongside it, then diff K3’s generated component against Fable 5’s on the identical ticket — the frontend axis where the arena ranked K3 first.

Watch list

• Fable 5 metered cutover (July 19 → 20) — the deadline is now tomorrow: Anthropic’s second extension keeps Fable 5 included at up to 50% of weekly limits on Pro, Max, Team, and eligible Enterprise seats through 11:59:59pm PT July 19; after that, access moves to usage credits at $10 / $50 per million input/output tokens. Anthropic frames the meter as temporary pending capacity, with no return date — and no third extension has appeared. Watch whether a last-minute reprieve lands before the 19th, and whether standard Enterprise seats can enable usage credits before the included allowance empties mid-session. (official, prior coverage)

  For anyone who’s wired Fable 5 into a daily workflow, tomorrow’s real question is cost: at $10/$50 per million, sessions that were “included” become metered line items, so it’s worth checking now whether your usage sits under the 50% weekly cap or spills into credits mid-session.

• Kimi K3 open weights (July 27) — Moonshot has released K3 via API but says the 2.8T weights go public July 27, reportedly under a modified-MIT-style license (unconfirmed until the model card ships). At that size, “open” is heavily gated by hardware — running it locally is out of reach for all but the largest rigs — so the near-term signal is less “download it” than whether hosted providers stand up K3 endpoints and whether the license carries the monthly-active-user clause Moonshot’s recent releases have used. Watch for the model card and the first independent, non-arena evals to land alongside the weights. (unconfirmed) (source)

  Realistically, the July 27 weights matter to most teams not as something to self-host but as downward pressure on hosted K3 pricing — the thing to watch is whether third-party inference providers stand up endpoints that undercut Moonshot’s own $3/$15, and whether the license’s rumored monthly-active-user clause limits who can.

• Gemini 3.5 Pro — delayed past July 17, no new date — with the July 16 delay report, the target that anchored this list for a week is gone and Google has not published a replacement; reporting frames it as a third missed deadline tied to coding shortfalls, with an upgraded Flash model in partner testing that could ship first. The headline specs — 2M-token context, Deep Think mode, pricing — remain reported, never confirmed, with no model card. The signal to watch is a Google post with an actual date and a published model card, not another leaked target. (unconfirmed) (report, prior coverage)

  Roadmaps that penciled in Gemini 3.5 Pro for a Q3 evaluation should now treat the leaked 2M-context and Deep Think specs as unusable planning inputs until a model card exists — a third missed deadline means the timeline itself, not just the date, is the unreliable part.

• MCP spec finalization (July 28) — the 2026-07-28 release candidate (stateless core, Extensions framework, Tasks, MCP Apps, authorization hardening) is in its validation window, with the Python/TypeScript/Go/C# SDKs updating against it in beta. The Streamable HTTP transport now requires the Mcp-Method/Mcp-Name routing headers that security analysts have flagged as fresh attack surface; list/read results also gain ttlMs/cacheScope caching fields. Watch for the remaining SDK support landing and any last header changes before the RC locks. (official, prior coverage)

  Teams already running a Streamable HTTP MCP server have one concrete prep before the RC locks: confirm your transport — and any proxy sitting in front of it — correctly handle, and don’t leak, the new Mcp-Method/Mcp-Name routing headers that analysts have flagged as fresh attack surface.