AI Coding Models in the Second Half of 2026
How to choose: GPT-6 and today's Claude flagships
As of 2026-10-19, the current flagship in OpenAI's official model catalog is GPT-6 Astra, and the starting point in Anthropic's official model overview is Claude Opus 5.5. The question is not which model is strongest, but which model fits your workload: early September 2026 brought Claude Fable 5.1 (09-01), Gemini 3.8 Flash (09-02) and GPT-6 Astra (09-03), and the top tier was rearranged. Below is a decision framework built around budget, task complexity and context needs, a snapshot of the models each vendor currently sells, and a map of the selection guides on this site.
Switch models on the same endpoint and measure the cost per token yourself; check /models for callable models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Updated 2026-10-22
The current flagship in OpenAI's official model catalog is GPT-6 Astra: OpenAI places GPT-6.1 Sol at near-Astra performance for complex work at a lower cost, and GPT-6 Luna is the efficient option for focused, high-volume tasks. Anthropic's official model overview starts most workloads with Claude Opus 5.5. The models callable on QCode include gpt-6-astra, gpt-6-sol, gpt-6.1-sol, gpt-5.6-sol, gpt-5.6-terra, claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5, glm-5.3, glm-5.3-flash, kimi-k3, deepseek-v4-pro and qwen3.8-max; QCode does not offer gpt-6-luna or gpt-5.6-luna, so please use another GPT model, such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra. Claude (Opus 5.5 / Opus 5 / Fable 5.1) and OpenAI (GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-5.6 Sol / Terra) are all live on QCode. In Codex sessions signed in with a ChatGPT account, gpt-5.4 was removed on 2026-08-31; the OpenAI API and API-key Codex sessions are not covered by that shutdown. The official releases after 2026-09-18: Anthropic's Opus 5.5 (2026-09-22) and Sonnet 5.5 (claude-sonnet-5-5, released 2026-09-28, available on QCode from 2026-09-29), OpenAI's GPT-6 Sol and Luna (2026-09-22) and gpt-6.1-sol (released 2026-09-29, callable on QCode from 2026-09-30), and xAI's Grok 4.7 (2026-09-21, QCode does not offer Grok). On the open-weight side DeepSeek has moved to deepseek-v4.1-flash (2026-09-10) and deepseek-v4-pro (GA 08-13, peak/off-peak pricing), while GLM-5.3 (08-14) and Kimi K3 (07-16, weights open-sourced 07-27) are on sale. Google's catalog now lists Gemini 3.8 Flash and 3.8 Live, while for Gemini 4 Pro only the start of pre-training on 2026-07-21 is confirmed — no date has ever been given.
State of play as of 2026-10-19.
Decision framework: a few questions first
Answering the questions below before you compare benchmark scores is more effective.
| Axis | Ask yourself | Lean toward | Typical scenario | Note |
|---|---|---|---|---|
| Budget sensitivity | Is this workload large enough that cost control matters? | Tight budget → deepseek-v4.1-flash / Claude Sonnet 5.5; room in the budget → GPT-6 Astra / Claude Fable 5.1 | Bulk completion, CI scripts, light classification | In OpenAI's official tier list Luna lists at about 1/25 of Sol (QCode does not offer Luna); DeepSeek's V4.1-Flash costs $0.15/$0.6 in the off-peak window and twice that at peak, so the spread matters most for large volumes; the higher the tier, the larger the reasoning overhead per call |
| Task complexity | Is this a long-horizon refactor or high-difficulty reasoning task, or routine feature work? | Hard tasks → GPT-6 Astra / Claude Fable 5.1; routine work → GPT-6.1 Sol / Claude Sonnet 5.5 | Cross-repo refactors, complex debugging, architecture decisions | The gap between flagship tiers shows up on the hardest tasks; for ordinary feature work a mid tier is better value |
| Context needs | How much context does a single task need? | GPT-6 Astra / GPT-6.1 Sol (1.05M) · Claude Fable 5.1 (1M) · Gemini 3.8 Flash | Long-document Q&A, large cross-file refactors | Context windows are generally in the 1M range; what separates the models is capability and unit price, not window size |
Where each vendor stands
Claude (Anthropic)
The current Anthropic lineup is Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5, and the official model overview starts most workloads with Claude Opus 5.5. Fable 5.1 is the current top tier (released 2026-09-01, $10/$50, 1M context, 128K output, reliable knowledge cutoff June 2026, Thinking is Adaptive and always on), kept for demanding reasoning and long-horizon agentic work. Opus 5.5 costs $4/$20, has 1M context, targets long-running agentic coding and knowledge work, and a cache hit costs 5% of the standard input price ($0.20); the previous Opus 5 lists at an official $5/$25, 1M context, and is still callable on QCode, but it is no longer the newest tier since Anthropic released Opus 5.5 (claude-opus-5-5) on 2026-09-22. Sonnet 5.5 costs $2/$10, the balance of speed and intelligence, was released on 2026-09-28, and claude-sonnet-5-5 has been open on QCode since 2026-09-29; the previous Sonnet 5 costs $2/$10 as well. Haiku 5.5 targets high-volume, latency-sensitive tasks such as classification, extraction and routing, at an official price of $0.10/$0.50 and up (a higher tier applies above 100K tokens); the previous Haiku 4.5 is $1/$5, the fastest tier.
OpenAI
The current flagship in the official model catalog is GPT-6 Astra, released 2026-09-03: OpenAI describes it as its most capable model for the most demanding work, model id gpt-6-astra, list price $10/$50, a 1.05M context, a 128K output limit including reasoning tokens, a knowledge cutoff of April 30, 2026 and reasoning tiers from low to max with no none. GPT-6.1 Sol (gpt-6.1-sol) sits at near-Astra performance for complex work at a lower cost, listed at $2/$10, with the same context and output limit; GPT-6 Luna (gpt-6-luna) is the efficient option for focused, high-volume tasks, and the official selection guidance pairs it with cost-sensitive, high-volume workloads. OpenAI's 2026-09-22 announcement says GPT-6 Sol and Luna are priced 50% below the promotional GPT-5.6 prices. The GPT-5.6 tiers: Sol (currently $4/$20 on the official pricing page, marked as a promotional price running at least through 2026-11-21), Terra (balanced, $2/$12) and Luna (the fastest tier; see the OpenAI pricing page for its unit price). GPT-5.6 Sol and Astra have identical context windows and output limits; the difference lies in the reasoning tier and the knowledge cutoff. There is also the approval-gated GPT-5.6-Cyber, which sits in the OpenAI Daybreak group (Daybreak Red tier) and needs separate approval and provisioning.
Gemini 3.8 Flash and 3.8 Flash Cyber shipped on 2026-09-02 at a published entry price of $0.75 input and $3.75 output per 1M tokens, marked as a promotional price through 2026-12-31, with $1.50/$7.50 from 2027-01-01; the Cyber build is limited to trusted defenders in the Fairwind program. The earlier 3.6 Flash (07-21) and 3.7 Flash (GA 08-13) are still there. Gemini 3.5 Pro is still not broadly available — the only verifiable official wording is that it is being tested with partners and will be released when it is ready; no date has ever been given, and the two rumored launch dates (July and 8-12) have both slipped, so do not treat them as fact. Whether a Gemini model is available is listed on /models.
Open-weight camp (DeepSeek / GLM / Kimi)
DeepSeek, GLM and Kimi are the workhorses of the open-weight camp. On the DeepSeek side there are deepseek-v4.1-flash (2026-09-10, called the smallest model of the new architecture family in the release notes, with native visual understanding; 1M / 384K, MIT, official off-peak price $0.15/$0.6) and deepseek-v4-pro (GA 2026-08-13, $0.435/$0.87 at launch). From 2026-09-10 deepseek-v4-flash and vision-exp were retired, and requests under those names are served by V4.1-Flash at the Flash price (changelog); the changelog entry of 2026-09-10 overturns the V4 Pro redirect and confirms that V4 Pro keeps being provided with unchanged billing. GLM-5.3 (2026-08-14) and Kimi K3 (2026-07-16) are still on sale, and Kimi K3's weights were open-sourced on 07-27. See the comparison at /deepseek-v4-1-flash-vs-v4-pro and the pricing at /deepseek-v4-1-flash-pricing.
Selection guides on this site
Below you find every new and updated selection article on this site, listed by topic from shutdown migrations and flagship comparisons to pricing guides.
GPT-5.4 Retirement Migration Guide
The shutdown date (8/31), what it covers and the replacement model
Fable 5 vs GPT-5.6 Sol
Specs, context and prices compared
How Are Mythos 5 and Fable 5 Related?
Which one everyday users can actually access
Claude Fable 5 Pricing Guide
Official rates and real subscription cost
Fable 5's 80.3% on SWE-Bench Pro
Where the score comes from and why it's contested
GPT-5.6 Sol / Terra / Luna: how to choose
Benchmark prices and scenarios per tier (refreshed)
GPT-5.6 Complete Guide
Benchmarks, pricing, and access
Gemini 3.5 Pro Release Tracker
What is confirmed and what is still a rumor (the 8/12 rumor has slipped)
GPT-5.6 Pricing Guide
Prices per tier, cache pricing and cost estimates
2026 AI Model Radar
Real-time availability status for every frontier model
GPT-5.4 Retirement Migration Guide
How the GPT-5.6 tiers divide the work, and which jobs are worth moving up to GPT-6 Astra.
Claude Fable 5 vs GPT-5.6 Sol
Specs, context and prices side by side; on the OpenAI side the flagship tier is now GPT-6 Astra.
What Is the Relationship Between Claude Mythos 5 and Fable 5?
How Mythos 5 and Fable 5 differ, and what Fable 5.1 and Opus 5.5 each cover in Anthropic's current model overview.
Recent Timeline
The key release milestones of the past few months.
The GPT-5.6 family opens up in full
Sol, Terra and Luna are open to everyone via API and Codex; QCode does not offer Luna (as of 2026-10-19).
GPT-5.6 price cuts
Luna was cut by 80%, Terra came down by 20% to $2/$12, and Sol stayed the same.
August's packed release week
GLM-5.3 released (same base as 5.2, post-training scaling, coding +50%); Qwen3.8-Max shipped on 08-03 and open-sourced Max-tier weights for the first time on 08-12; GPT-5.6-Cyber moved into the OpenAI Daybreak group on 08-10 and needs separate approval and provisioning.
Three vendors in one week, top tier rearranged
Anthropic released Claude Fable 5.1 and Mythos 5.1 on 09-01; Google shipped Gemini 3.8 Flash and 3.8 Flash Cyber on 09-02; OpenAI released GPT-6 Astra on 09-03. Astra and Fable 5.1 carry exactly the same list price ($10/$50) with close context and output limits, a same-price confrontation across the camps in the second half of 2026. The top tiers of Claude and OpenAI ended up in the same price band that week.
DeepSeek changes gear: V4.1-Flash ships, the 0731 build retires
On 2026-09-10 the official release shipped V4.1-Flash and cut the price of the Flash tier, while retiring deepseek-v4-flash and deepseek-v4-flash-vision-exp; the changelog entry of the same day overturned the V4 Pro redirect. Around the same time the official API catalog also listed Gemini 3.8 Live (id gemini-3.8-live).
Where the information comes from
The release dates, model positioning and list prices on this page come from OpenAI's official model catalog and API pricing page, Anthropic's official model overview and pricing page, the OpenCode Console model list and price table, and qcode.cc/models. The OpenAI and Anthropic entries were checked line by line on 2026-10-19; Google's model catalog and the DeepSeek news260910 release note / changelog keep the results of the September 2026 check, with the capture dated 2026-09-18 and the entries added between 2026-09-21 and 2026-09-29 checked on 2026-09-29.
FAQ
What's the first thing to check when picking an AI coding model today?
Don't jump straight to benchmark scores. Answer a few questions first: budget sensitivity, task complexity, and per-task context needs. Once those are answered, the candidate list usually narrows to a handful of tiers, and benchmark scores can settle the final call. To try a new model for a few days first and not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days (top-ups start at $5).
Do all GPT-5.4 users need to migrate?
No. Officially, such shutdowns only cover Codex sessions signed in with a ChatGPT account: gpt-5.4 was switched off on 2026-08-31 (re-checked on 2026-08-30, still "starting August 31"), and GPT-5.5 retired from Codex with ChatGPT sign-in on October 14, 2026, at noon, with GPT-6.1 Sol named as the replacement for signed-in users. The OpenAI API and Codex sessions authenticated with an API key are not covered by this announcement; on the OpenAI API you can keep using them, or move to the matching GPT-5.6 tier.
Is Gemini 3.5 Pro available yet?
Not yet. As of 2026-08-22, Gemini 3.5 Pro was still in a limited preview; the only verifiable official wording is that it is being tested with partners, both rumored launch dates (July and August 12) have slipped, and Google has not confirmed any new date — treat it as a rumor, not as a basis for decisions.
How is this page different from the 'AI Model Radar'?
The radar page lists the current state of each model: which are available, which are in preview, which are restricted. The selection side gives you the criteria — budget sensitivity, task complexity, per-task context needs — and collects the new and updated selection articles on this site. The radar page shows the status quo, the selection content shows how to choose.
Create a QCode account
One key across Claude, GPT and Chinese models, billed per token — live rates on /models.
Related Reading
2026 AI Model Radar
Live availability tracking for every frontier model, which pairs well with the decision framework above.
Claude Fable 5 on SWE-Bench Pro
Where the 80.3% score comes from, and what it's actually worth as a signal.
Claude Fable 5 Pricing Guide
Official rates, cache pricing, and real subscription-vs-API cost.
Prices, benchmarks and release dates on this page were cross-checked: the OpenAI and Anthropic entries on 2026-10-19, the Google and DeepSeek entries keep the September 2026 results. The release date of Gemini 3.5 Pro is still an unconfirmed rumor. Information can change at any time; model availability is listed on /models, and the official announcements from each vendor are authoritative.