Fleet Radar — 2026-11-06

_Synth: deepseek/deepseek-v4-pro · Sweep: perplexity/sonar-pro · 453 models live · +3 new / ~13 changed / -5 gone_

What changed today

• New models: google/gemini-omni-1.2-flash (video generation, 131k ctx, text+image+video → text+video); google/gemini-omni-1.1-flash (superseded, will be routed to 1.2 on Dec 3); openrouter/auto-code (2000k ctx, unparsable $-1000000 pricing – likely non‑functional)
• Notable price spikes (output per 1M tokens):

- deepseek/deepseek-v4.1-flash: $0.051 → $0.567 (11×) – directly hits dsf tier

- deepseek/deepseek-v4-pro: $0.897 → $1.795 (doubled) – hits ds tier

- deepseek/deepseek-v4-pro-0813: $0.117 → $4.66 (40×) – likely a premium variant

- moonshotai/kimi-latest: $0.182 → $10.152 (56×) – kimi‑code alert

- Others up: deepseek/deepseek-chat, minimax/minimax-m2.7, nvidia/nemotron-3-super-120b-a12b, google/gemma-4-26b-a4b-it, etc.

• Many models also gained maxOut (higher output token limit): reflection/beam-1.0-501b-a23b, deepseek/deepseek-v4-flash-vision-exp, deepseek/deepseek-v4-pro, google/gemma-4-26b-a4b-it, minimax/minimax-m2.7, nvidia/nemotron-3-super-120b-a12b, deepseek/deepseek-chat, meta-llama/llama-3.3-70b-instruct

Off‑catalog intel (web sweep)

• Google Gemini Omni 1.2 Flash GA with 15‑second video generation, improved reference consistency; video output pricing dropped 30% to $12/1M tokens. (sweep source 1, 2)
• OpenAI GPT‑6 Cyber available to Daybreak Red customers for vulnerability investigation; GPT‑6 Rosalind added to catalog. (sweep source 3)
• Kimi K3.1 Pro (1M‑ctx, vision, tool calls, structured output) listed on serverless at $2/$10/$0.20 per million tokens. (sweep source 4)
• Anthropic launched Claude Console IP allowlist beta and Personal Finance in Claude – no new model. (sweep source 5)
• Mistral Large 4 announced as 1T‑param multimodal MoE with 1M context, open weights – unverified primary source. (sweep source 6)

New capabilities & features

• Higher output ceilings (maxOut) enabled on the models listed above under “changed”.
• Google Gemini Omni 1.2 Flash adds native video generation (text+image+video → text+video output) – see Omni Flash guide.
• OpenAI Codex now serves gpt-6-cyber for cybersecurity workflows (Daybreak setup required).
• Kimi K3.1 Pro (off‑catalog) brings tool‑use, structured outputs, and 1M‑context reasoning with vision – capable of messaging API for agentic use.

Best in class right now

• agentic‑terminal‑coding – codex tier (openai/gpt-5.5-xhigh), still unchallenged for Codex‑integrated terminal coding.
• algorithmic‑coding – ds tier (deepseek/deepseek-v4-pro), despite doubled price; DeepSeek V4 Pro remains top for math/code reasoning.
• hard‑reasoning – claude tier (Opus‑tier), still the strongest for multi‑step, high‑complexity logic.
• long‑context (>=512k) – agy/pro tier (google/gemini-3.1-pro), 1M+ window with reliable retrieval; DeepSeek’s 1M ctx is close but Gemini edges quality.
• vision / multimodal – agy/vision tier (google/gemini-3.1-pro-vision) for still‑image understanding; for video generation, google/gemini-omni-1.2-flash is now best.
• image‑generation – google/gemini-omni-1.2-flash (video generation); for still‑image generation, no dedicated fleet leader today – could be routed to agy/vision using Omni Flash.
• embeddings – no change; a free embedding model (e.g., nvidia/nemotron-3-nano-8b on free*) remains adequate for lightweight tasks.
• cheap‑bulk‑generation – agy/flash tier (Gemini Flash) under AI‑Pro quota gives zero marginal cost with strong quality; if quota‑exhausted, free* (NVIDIA Nemotron 3 free tier) is truly zero‑cost.
• local / on‑device – local/fm (Apple on‑device) or local (Qwen3.5‑9B MLX) remain unrivaled.
• best‑value (capability per $) – free* tier for zero‑cost bulk; for low‑cost multimodal with context, google/gemma-4-26b-a4b-it at $0.3/M output offers strong value.

Proposed route‑table changes

• dsf tier: The current model deepseek/deepseek-v4.1-flash saw an 11× price jump. Switch dsf to the unchanged deepseek/deepseek-v4-flash (if available and cheaper), or reroute bulk reasoning to agy/flash (Gemini Flash) leveraging AI‑Pro quota – it is effectively free within quota and fast.
• ds tier: Verify that ds is not using deepseek/deepseek-v4-pro-0813 (priced at $4.66/M); stick to deepseek/deepseek-v4-pro at $1.795/M (still reasonable).
• kimi‑code tier: The huge price jump of moonshotai/kimi-latest may make it untenable. Consider evaluating Kimi K3.1 Pro (off‑catalog, competitive pricing) as a replacement if it appears on OpenRouter, or fall back to other coding‑agent tiers.
• agy/vision tier: Add google/gemini-omni-1.2-flash as the video‑generation endpoint; keep google/gemini-3.1-pro for image understanding.

Gaps & uncertainty

• No benchmark data for Kimi K3.1 Pro, Mistral Large 4, GPT‑6 Rosalind, or Ripple‑65B – performance vs. existing fleet leaders is unverified.
• openrouter/auto-code pricing is garbled; model may not be operational.
• Several price hikes (esp. deepseek/deepseek-v4-pro-0813 and kimi-latest) could be transient or catalog errors – need confirmation from providers.
• Fleet quotas (AI‑Pro, ChatGPT Plus) and actual usage limits are unknown; agy/flash bulk feasibility depends on quota headroom.
• Off‑catalog Mistral Large 4 (open weights) could be a best‑value contender but no API availability yet.

_No gaps: every source answered._

Fleet Radar · live data, never training · 2026-11-06