Fleet Radar — 2026-07-07

_Synth: deepseek/deepseek-v4-pro · 343 models live · +0 new / ~0 changed / -0 gone_

What changed today

No new models or material shifts detected; landscape is unchanged.

No new capabilities or features detected today (no features:+ tags, changelog excerpts unreachable).

agentic-terminal-coding: openai/gpt-5.5-xhigh (codex) — dedicated agentic-coding leader with high reasoning and tool use.
algorithmic-coding: deepseek/deepseek-v4-pro (ds) — top performer on algorithmic benchmarks, cost-effective reasoning.
hard-reasoning: google/gemini-pro (agy/pro) — best hard-reasoning performance, 2M context, free via AI-Pro quota.
long-context (>=512k): google/gemini-pro (agy/pro) — up to 2M token window, strong retrieval and reasoning.
vision / multimodal: google/gemini-3.1-pro (agy/vision) — best-in-fleet image understanding, natively multimodal.
image-generation: N/A — no fleet model; recommend external (e.g., dall-e-4 or Midjourney v7).
embeddings: N/A — no fleet embedding model; use text-embedding-3-large or voyage-3-lite externally.
cheap-bulk-generation: nvidia/nemotron-3-ultra (free*) — zero-cost via OpenRouter free tier; also google/gemini-flash (agy/flash) for slightly better quality at minimal cost.
local / on-device: qwen/qwen3.5-9b-mlx (local) — best on-device codegen and reasoning; Apple on-device model (local/fm) for strictly offline/sensitive tasks.
best-value: deepseek/deepseek-v4-pro (ds) — top code/reasoning capability per dollar.

None.

LIVE data missing provider changelog excerpts; all capability rankings based on fleet descriptions and yesterday’s known positions.
Exact model IDs for agy/pro (google/gemini-pro vs. gemini-2.5-pro) and agy/vision (google/gemini-3.1-pro) are inferred; no direct benchmark verification.
Image-generation and embeddings buckets have no fleet representatives—recommendations are external only.

_No gaps: every source answered._

Fleet Radar · live data, never training · 2026-07-07