Best models by lab — current flagships
## what it does
tl;dr
Flagship names move every few weeks. As of 6 Sep 2026 (this session’s fetch): OpenAI GPT-6 Astra, Anthropic Claude Fable 5.1 (included usage only on Max / premium seats, cap 50% of weekly limit), Google Gemini 3.8 Flash, xAI Grok 4.6 (grok-4.6). Free tiers do not get the new OpenAI frontier. Re-read the sources in config/feeds.toml before you trust last month’s slug.
This page is a changelog digest, not a sandbox bake-off. sandbox_test.verdict stays untested.
How to keep this current
Pull these on each rewrite (all in feeds.toml):
| Lab | What to fetch |
|---|---|
| OpenAI | GPT-6 Astra post, news RSS, API models |
| Anthropic | News, Fable 5.1, API release notes |
| AI RSS, 3.8 Flash post | |
| xAI | Grok 4.6 post, models + token prices — not x.ai/pricing (403 from this IP) |
| Cross-check | OpenRouter models, LMArena |
| Copilot | GitHub changelog RSS — Astra showed up in Copilot the same week |
Independent arenas lag launches. If OpenRouter does not list the API id yet, the consumer app may still have it.
Snapshot — 6 Sep 2026
| Lab | Use this | Also shipping | Paid gate (from vendor copy) |
|---|---|---|---|
| OpenAI | GPT-6 Astra | GPT-5.6 Sol / Luna still on the pricing grid | chatgpt.com/pricing: Plus $20 includes GPT-6 Astra. Pro from $100 (5×) or $200 (20×) = more usage + Pro reasoning on Astra. Free = Luna, not Astra. |
| Anthropic | Claude Fable 5.1 (claude-fable-5-1) | Claude Opus 5 still on the platform; Mythos 5.1 = same weights, trusted-access safeguards | Not “included on every paid plan.” Help Center: included on Max and premium Team/Enterprise seats for up to 50% of weekly usage. Pro and standard Team seats run Fable on usage credits (extra). Free: no Fable. Max web prices: $100 / $200 (5x / 20x). |
| Gemini 3.8 Flash | 3.8 Flash Cyber via Fairwind (defenders). Gemini 3.1 Pro is the last Pro-card we fetched (Feb 2026). | Consumer 3.8 Flash: Google AI Pro and Ultra. API listed at intro $0.75 / $3.75 per 1M tokens until 31 Dec 2026. | |
| xAI | Grok 4.6 (grok-4.6) | 4.5 / 4.3 / grok-build still listed on the same docs table | Docs: “for everything else, including code, use Grok 4.6.” Token price $2 / $6 per 1M under 200k prompt; $4 / $12 if prompt ≥ 200k. Fast variant 2×. Consumer SuperGrok HTML still 403 here. |
OpenAI’s own Astra post (not a recap blog): biggest published jumps vs GPT-5.6 Sol on computer use, Terminal-Bench 4.0 (57.9% vs 37.3%), Terminal-Bench Science (64.6% vs Sol 22.4% in their chart), and they call it a step change on ARC-AGI-3. Treat those as vendor evals.
Anthropic’s Fable 5.1 table vs Fable 5 / Opus 5 / GPT-5.6 Sol: Terminal-Bench-Science 52.6% (Fable 5.1) vs 22.4% (Sol) vs 29.0% (Opus 5). CursorBench 3.2.0: 73.4% vs Sol 67.2%.
Claude Fable — plan facts (Help Center)
Source: Claude Fable models on your plan. Same rules for Fable 5 and 5.1. Max prices: What is the Max plan? — Max 5x $100/mo, Max 20x $200/mo. Plan ladder: Choose a Claude plan (Pro $20/mo).
| Plan | Fable 5 / 5.1 |
|---|---|
| Free | Not available |
| Pro ($20) | Available, not in weekly included limits — pay-as-you-go usage credits from the first Fable message. No 5.1 promo credit (July credit was Fable 5 only). |
| Max 5x ($100) / Max 20x ($200) | Included. Up to 50% of the weekly usage limit on Fable, at no extra cost. Same weekly bucket as other models (not 50% extra). Fable burns the limit faster. After 50%, credits or switch model. |
| Team / Enterprise premium seats | Same as Max: included, 50% cap. |
| Team / Enterprise standard seats | Same as Pro: credits, not included. |
| API | Standard API rates (Fable list was $10 / $50 per 1M in the Fable 5.1 product post; cache reads cheaper vs Fable 5). |
Calling Fable “generally available” without this table is wrong. The model exists on Pro; included quota starts at Max ($100+).
xAI Grok 4.6 (Aug 12 post + model card): 500k context, knowledge cutoff 1 Feb 2026, in Cursor / Grok Build / API / OpenRouter.
GPT-6 Astra — what actually shipped
Source: openai.com/index/gpt-6-astra.
- Rollout: chatgpt.com/pricing now lists GPT-6 Astra on Plus ($20/mo). Pro is 5× ($100) or 20× ($200) usage vs Plus, with Pro reasoning on Astra.
- Free / Go: not the Astra row on that page (Free is GPT-5.6 Luna).
- API:
gpt-6-astra. Standard $10 / $50 per 1M input/output. Fast mode = 2× speed at 2× Standard price. - Also Azure and AWS Bedrock.
- Codex: experimental notes across context windows (searchable prior windows instead of only compaction). Config: Codex config.toml. Default-on for Astra “in the coming weeks” per the post.
GitHub’s changelog the same week listed GPT-6 Astra in Copilot. That is a second product; plan mapping is GitHub’s, not ChatGPT’s.
If someone says “only Pro has Astra”: Pro reasoning / Astra Pro, yes. Base Astra is on Plus on the live pricing page. Free still does not get it.
Pick by job (this week)
| Job | First try | Why |
|---|---|---|
| Agentic coding / computer use | Astra or Fable 5.1 (Fable only if Max/credits) | Both labs published large TB / computer-use jumps this week. |
| Long Claude Code sessions | Fable 5.1 on Max / premium or usage credits | High effort default in Claude Code. Pro $20 does not include Fable in the weekly bucket. |
| xAI / Cursor | Grok 4.6 | Current xAI text flagship. API $2/$6 per 1M (<200k prompt). |
| Cheap high-volume agents | Gemini 3.8 Flash | Flash cadence; Pro-class Gemini may still be the older 3.1 Pro card. |
| Stay on free | Local (Ollama, LM Studio) | You will not get Astra-class computer use or included Fable. See pay for AI. |
Related
## embed this badge
