Updated September 2026 · live status

The 2026 AI Model Radar

GPT-6 Astra, Claude Fable 5.1, DeepSeek V4.1 Flash, Gemini 3.8 Flash and more — which models are live, which (such as Gemini 4 Pro) are still rumour, which are gated, and how to be first on them when they open.

Updated 2026-09-20

#Release Tracker #Availability #Benchmarks #How to Access
Status legend
Available now Limited preview Suspended Coming soon Evaluating

Frontier model status board

Where each major 2026 model stands today — click any card for the full guide.

Anthropic · Claude

OpenAI · GPT / Codex

Google · Gemini

Gemini 3.5 ProDelayed

Absent from the July 21 triple launch. Google said it fell short of internal expectations on coding and complex reasoning, so the broad release is delayed; it remains in partner testing.

Gemini 3.5 FlashAvailable now

GA'd at Google I/O on May 19, 2026 — fast and multimodal, usable via Gemini on QCode.

Gemini 3 ProAvailable now

1M-context flagship from November 2025 (LMArena 1501). Available via Gemini on QCode.

Gemini 3.6 FlashAvailable now

Shipped 2026-07-21 as gemini-3.6-flash: $1.50/$7.50 Paid, 1,048,576 / 65,536 context; stronger coding/multimodal workhorse vs prior Flash.

Gemini 3.5 Flash-LiteAvailable now

Shipped in the same batch as the most cost-effective model in its class.

Gemini 3.5 Flash CyberAvailable now

Also from the July 21 batch: a security-specialised model tuned for finding and fixing vulnerabilities.

Gemini 4Rumored

Google confirmed on 2026-07-21 that Gemini 4 pre-training had begun, but published no date, specifications or Pro-tier SKU.

xAI · Grok

Grok 4.5Evaluating

Tops AutomationBench-AA at 51.4% (Fable 5 48.6%, Opus 4.8 48.5%) and ranks 4th on the Artificial Analysis Intelligence Index at roughly $0.34 per task. The trade-off is hallucination rising from 25% to 54% (AA-Omniscience). QCode does not currently carry the Grok family.

Grok BuildEvaluating

xAI's build-oriented agent environment; see the dedicated page for its open-source ecosystem and toolchain. QCode does not currently carry the Grok family.

Grok 4.6Evaluating

Released 2026-08-12. Artificial Analysis Intelligence Index 61, level with GPT-5.6 Sol; $2/$6 with a 500K context and $0.5 cache hits. QCode does not currently carry the Grok family.

Grok BotLimited preview

In early beta since 2026-08-11: a "computer teammate" on a shared cloud desktop, bundled with Cursor Ultra ($200/mo) or Teams Premium ($120/seat/mo). A third-party product, not a model endpoint.

Grok 4.7Rumored

Listed on docs.x.ai (checked 2026-09-22): grok-4.7, 500k context, $2.00 in / $6.00 out per million tokens, reasoning configurable; no official benchmarks yet. The founder's 2.1T figure is unconfirmed. Slip timeline (09-02 / 09-11 / 09-14): /grok-4-7-status-tracker.

Claude Opus 5.2Rumored

The live Opus row in the official table is still claude-opus-5; neither “Opus 5.1” nor “5.2” has one, and seeing a slug in a console is not release evidence. Tracker at /claude-opus-5-2-status-tracker.

Claude Fable 5.2Rumored

Anthropic's comparison table currently lists one Fable row, claude-fable-5-1 (2026-09-01), while claude-fable-5 sits in the Legacy list on the same page; 5.2 has no row, no id, no date, no price. Tracker at /claude-fable-5-2-status-tracker.

GPT-6 SolAvailable now

Officially released by OpenAI on 2026-09-22, in the same announcement as GPT-6 Luna. Official id gpt-6-sol, $2/$10, cached input $0.20, 1,050,000 context, 128,000 max output; calculated from the official price table, half of gpt-5.6-sol ($4/$20). Callable on QCode. Tracker at /gpt-6-sol-status-tracker.

Gemini 3.8 Flash / LiveAvailable now

Google's API directory as of 2026-09-17 lists Gemini 3.8 Flash and 3.8 Live (id gemini-3.8-live) together; Google published no separate GA dates, so this card does not invent them. Our /models lists gemini-3.8-flash and gemini-3.8-live; guide at /gemini-3-8-flash-guide.

Claude Fable 5.1Available now

Anthropic's model table lists releasedOn 2026-09-01 with lifecycle active — the live build of the first Fable. On QCode as claude-fable-5-1; guide at /claude-fable-5-1-guide.

Chinese open-weight models

GLM-5.2Available now

One of the strongest open-weight coding models (MIT, released June 2026, 744B with 40B active, 1M context). Live on QCode as glm-5.2 and glm-5.1.

DeepSeek V4Available now

The two live tiers of the V4 family today are deepseek-v4.1-flash (2026-09-10) and deepseek-v4-pro (GA 2026-08-13), both MIT weights with 1M context. Both ids sit in our /models with real 30-day traffic.

DeepSeek V4 Flash 0731Available now

The 2026-07-31 post-training update: MIT weights, 1M / 384K, $0.14/$0.28 at the time, Terminal-Bench 2.1 82.7. On 2026-09-10 DeepSeek shipped V4.1-Flash and retired deepseek-v4-flash, whose requests are now served by deepseek-v4.1-flash. History page /deepseek-v4-flash-0731.

DeepSeek V4 Pro 0813Available now

GA on 2026-08-13, 1.7T params, MIT weights, $0.435/$0.87 at launch, Terminal-Bench 2.1 87.9, Artificial Analysis intelligence index 53. The changelog's 2026-09-10 entry confirms V4 Pro stays available with billing unchanged. On QCode as deepseek-v4-pro.

GLM-5.3Released · not in our catalog

Released 2026-08-14 (same base as 5.2, post-training scaling, coding +50%, tops CyberGym at 84.5%); API went live 08-19. QCode /models already lists glm-5.3 ($1.40/$4.40) with real calls over the last 30 days. Details: /glm-5-3-guide. The same-generation low-cost tier, GLM-5.3-Flash, shipped 08-26 (320B-A18B, 1M context, MIT open weights, promo $0.075/$0.25 through 09-09). Not connected on QCode; reference: /glm-5-3-flash-guide.

Kimi · MiniMax · QwenAvailable now

Kimi K3 and K2.6 plus the Qwen 3.7 family are now live on QCode; MiniMax and others remain under evaluation.

DeepSeek V4.1 ProRumored

DeepSeek named V4.1-Pro once, in the 2026-09-10 release note, and the changelog's 2026-09-10 entry withdrew the V4 Pro redirect it belonged to: no date, no id, no price. Tracker on /deepseek-v4-1-pro-status-tracker.

DeepSeek V4.1 FlashAvailable now

Released 2026-09-10: the note calls it “the smallest model in our new architecture family”, with native visual understanding. On QCode as deepseek-v4.1-flash, 1M / 384K, two peak/off-peak rates. Guide at /deepseek-v4-1-flash-guide.

Status as of 2026-09-18. Availability and benchmarks move fast; preview, delayed and rumoured entries are labelled.

✅ What you can use on QCode right now

One API key, three platforms — pick the models that are actually live today.

Claude Fable 5.1

Anthropic's table marks it releasedOn 2026-09-01, lifecycle active. Sold on QCode as claude-fable-5-1.

GPT-6 Astra

Shipped 2026-09-03 as gpt-6-astra, list $10/$50. Codex calls that id directly; it carries real 30-day traffic here.

DeepSeek V4.1 Flash and Gemini 3.8 Flash

Both are in our /models: deepseek-v4.1-flash (2026-09-10, which DeepSeek calls the smallest model in its new architecture family) and gemini-3.8-flash.

How to get new models the moment they open

Create a QCode account — signing up itself costs nothing. While your plan is active, we switch new models on for you the instant they land: no extra sign-up, no key swap. The same balance and API key work across Claude, Codex and Gemini.

Frequently asked questions

Can I use GPT-5.6, Fable 5 or Gemini 3.5 Pro on QCode today?

GPT-5.6: yes — it went GA on July 9 and QCode has onboarded Sol / Terra / Luna. Claude Fable 5: yes on the API (our route reopened July 15), while Anthropic's own subscriptions have tiered it by plan since July 19. Gemini 3.5 Pro is still in enterprise preview — we will open it the moment it is released. Claude Opus 4.8, Codex (GPT-5.6) and Gemini 3 Pro / 3.5 Flash are live as well.

Why was Claude Fable 5 taken offline?

Two separate things. First, upstream availability: Anthropic suspended Fable 5 and Mythos 5 worldwide on June 12, 2026 to comply with a US export-control directive (other Claude models were unaffected); Fable 5 access was restored on July 15 and it is available again on QCode, while Mythos 5 remains limited to approved organizations. Second, subscription rules: since July 19 Anthropic has tiered Fable 5 on its own plans — standard on Max and on Team/Enterprise premium seats, pay-as-you-go usage credits on Pro and standard Team seats. The API and consumption-based Enterprise are not affected by the second one.

Which 2026 model is best for coding?

Among generally available models, GPT-5.6 Sol leads terminal agentic coding at 88.8% on Terminal-Bench 2.1, while Claude Opus 4.8 is still the usual yardstick for deep refactors and judgment-heavy tasks. GLM-5.2 is among the strongest open-weight models, and QCode /models lists glm-5.2. We keep this radar updated as benchmarks land.

Does QCode add new models automatically?

Yes. When a model becomes broadly available and we secure capacity, we enable it for all accounts — your existing key and balance work immediately.

Be first in line for every 2026 launch

Activate a plan from $8.57/mo and use Claude, Codex and Gemini today — new models switch on for the same key the moment they open.

Try first, then decide

Not sure which tier? Start with Starter ($8.57/mo) and upgrade when you're happy — the unused value of the old plan goes back to your balance.