brain/
sourceartificial-intelligence

X overnight: OpenCode Omen Alpha, Unsloth GLM-5.3-Flash speedup, Astra Foundry

Overnight/morning X pass: OpenCode stealth Omen Alpha for Go subs, Unsloth Sep 4 local speedup for GLM-5.3-Flash, and Astra messy-rollout / Azure Foundry follow-through.

Source

X overnight: OpenCode Omen Alpha, Unsloth GLM-5.3-Flash speedup, Astra Foundry

Generated by Grok Bot research on 2026-09-04. Built-in WebSearch/WebFetch + native X (no Parallel). Treat as raw material — review before promoting into a project or thread.

Summary

Overnight into the morning of 2026-09-04, three AI signals showed up on X that were not in yesterday’s AI-thread X sources. OpenCode announced a new stealth coding model, Omen Alpha, exclusive to OpenCode Go subscribers and pitched as $100 usage for $10. Community fingerprinting and secondary press already route it toward Zhipu/GLM (same pattern as Ox Alpha → GLM-5.3-Flash), but Z.ai posted nothing overnight confirming Omen. Separately, Unsloth published a Sep 4 local-inference speedup for the already-known GLM-5.3-Flash / Ox Alpha open weights. On the closed-frontier side, OpenAI’s messy GPT-6 Astra ChatGPT rollout continued overnight with staff banked-reset messaging, while Satya Nadella pointed at a Microsoft Azure blog placing Astra in the Foundry Limited Access Program.

Findings

OpenCode launches stealth model Omen Alpha (Go exclusive)

At 2026-09-04 05:29 UTC, verified @opencode posted: “Omen Alpha (new stealth model) / Exclusively for OpenCode Go subscribers / $100 usage for $10.” The post has a video attachment; claims here are taken from the post text only (not transcribed). The same morning, OpenCode’s Go marketing page listed Omen Alpha in the paid Go lineup alongside other open-weight coding models and stated Go is a $10/month subscription (OpenCode Go).

A live data URL that community posts used to argue Zhipu provenance — opencode.ai/data/zhipu/omen-alpha — returned no model facts or usage rows when fetched on 2026-09-04 morning (“No model data”). Do not treat that path as confirming a live Zhipu attribution table at fetch time.

Fingerprinting / press: Omen as another Zhipu stealth (unconfirmed by Z.ai overnight)

Discourse posts (not lab-official) argued Omen is a GLM-family drop:

  • @imnotchalk (2026-09-04): “Omen Alpha is a GLM model” pointing at the OpenCode data URL above.
  • @imnotchalk (2026-09-04): “almost certainly a new GLM model… Most likely GLM 5.4 or GLM 5.4 Flash,” plus a claimed tokenizer match vs recent GLM models (screenshots; not independently re-run here).
  • Benchmark-style video posts by @SPAC89 comparing Omen runtime to GLM-5.3 Flash / GPT-5.6 Luna Max are video-dependent anecdote — flag only; do not promote numbers from them until transcribed.

Secondary press Lookonchain / Dongcha flash (fetched 2026-09-04) repeats the Go $10 / $100 usage framing, adds claimed per-token rates ($0.20 / $0.66 per million in/out), and states Zhipu has not officially acknowledged Omen. Treat Lookonchain token rates as secondary until OpenCode or Z.ai publishes a model card. get_users_posts on @Zai_org for the overnight window returned zero original posts.

Prior vault entity glm-5-3-flash already files the Aug 2026 Ox Alpha = GLM-5.3-Flash identity from lab-adjacent X; this overnight clip does not silently rename Omen to a GLM SKU.

Unsloth: Sep 4 local speedup for GLM-5.3-Flash (Ox Alpha)

Verified @UnslothAI (2026-09-04 12:32 UTC) posted that GLM-5.3-Flash now runs “3.3x faster locally,” with “Local GGUF inference… 1.6–3.4× faster with optimized decoding and bonus multi-token prediction,” pointing to Unsloth docs and Hugging Face GGUFs.

Issuer docs on unsloth.ai/docs/models/glm-5.3-flash (fetched 2026-09-04) match and expand those claims: banner “Sep 4: GLM-5.3-Flash now runs with 3.3x faster inference”; model described as also known as ox-alpha, 320B total / 18B active multimodal open model; 1M context; hardware table (1-bit ~100 GB through BF16 ~650 GB); tok/s tables for optimized decoding and MTP (e.g. tg32 @ 65536: 20.66 → 48.99 tok/s before MTP n=2 examples). GGUF hub listing: huggingface.co/unsloth/GLM-5.3-Flash-GGUF.

Separately, @BAI_AGI claimed GLM-5.3-Flash (Ox Alpha) hit #1 on B.AI with “2.41 trillion” cumulative tokens and restated 320B/18B / 1M context — treat as platform marketing, not a Z.ai model card.

GPT-6 Astra overnight: messy ChatGPT access + Azure Foundry Limited Access

Yesterday’s AI-thread source already files the public Astra rollout. Overnight additions:

  • OpenAI staff @thsottiaux (2026-09-03 23:12 UTC): “one banked reset for every day you don't have access to Astra on your paid ChatGPT plan, starting today.”
  • @sama (2026-09-04 01:02 UTC): apologized for the “messy rollout,” said they “should be able to begin broad rollout to API customers and chatgpt subscribers in the near future,” starting with Pro.
  • @satyanadella (2026-09-04 03:21 UTC) linked the Azure blog; @sama quote-posted “We are also excited!”

Issuer Azure blog (dated September 3; fetched 2026-09-04): Astra “begins rolling out today through the Microsoft Foundry Limited Access Program, with availability expanding to participating customers over the coming days.” Emphasizes multi-step planning, polished artifacts, and computer use across applications, with Foundry identity/governance controls; notes OpenAI reports SOTA on selected computer-use evals and that prompts/outputs are not used to train the models. Schema.org on the page listed datePublished 2026-09-03T18:15:00+00:00.

Official @OpenAI, @GoogleDeepMind, @AnthropicAI, @karpathy, and @GeminiApp timelines returned zero original posts in the overnight window (start_time 2026-09-03T22:00:00Z). @xai timeline was not authorized for this OAuth context.

Contradictions and open questions

  • Omen identity: OpenCode confirms the stealth name and Go exclusivity; Z.ai has not confirmed which GLM (if any). Community “GLM 5.4 / Flash” labels are speculation.
  • OpenCode data URL: Community screenshots vs this morning’s empty zhipu/omen-alpha data page — attribution may have been transient or path-dependent.
  • Lookonchain token prices ($0.20 / $0.66) are not on the @opencode announce text or the Go marketing page fetch; keep secondary until an issuer table lands.
  • Astra access: Staff banked-reset + “near future” broad rollout vs Azure “Limited Access Program… coming days” — channels differ; do not collapse ChatGPT consumer access with Foundry enterprise access.
  • Video posts (@opencode announce promo; @SPAC89 benchmark clips; @BAI_AGI promo) were not transcribed; no video-only claims promoted.

Provenance

Method: Grok Bot / WebSearch + WebFetch + curl ladder / native X (search_news pointers only, then get_users_posts / get_posts_by_id(s); skipped search_posts_all on user-OAuth) Generated: 2026-09-04 fetch_method: x-mcp Window: overnight/morning before weekday 10am X ingest; lab accounts quiet; news clusters used only as post-id pointers (not cited as grain)

Web sources:

X sources:

Grokipedia:

  • not used
Referenced by
Concepts
Chinese open-weight models approach frontier parityFrontier-class intelligence is escaping the datacenter to the edgeOpenAI Astra crossed Preparedness Framework "Critical" cyberThe US government is now in the frontier-model release loop
Entities
GLM-5.3 Flash / Ox AlphaGPT-6 AstraLyria 3.5MicrosoftOmen AlphaOpenAIOpenCodeUnsloth