2026 AI Coding Latest Updates
As of 2026-11-05, the newest dated official updates on the tooling and API side: on 2026-11-04 Antigravity CLI 1.5.0 added acp to agy --help — agy acp runs the CLI as an Agent Client Protocol (ACP) server over stdio so editors can use it as their agent; on 2026-11-04 Claude Code 2.1.318 added the date a stored claude.ai sign-in ends to claude auth status (loginExpiresAt) and to claude -p JSON results (credential_expires_at), plus a one-line stderr warning in claude -p text output starting 10 days before it ends; and on 2026-11-02 Anthropic put message threads for the Messages API into beta (beta header message-threads-2026-08-12): the API keeps the conversation on the server so each later request sends only the new messages, and a thread expires 5 minutes after its last request and is deleted no more than 1 hour after its last use. As of 2026-10-13 the most recent official release is Anthropic's Claude Haiku 5.5 (claude-haiku-5-5) on 2026-10-07, which Anthropic calls the lowest-priced, fastest and most capable small model it has ever released: on the Claude Platform it costs $0.10 per 1M input tokens and $0.50 per 1M output tokens for prompts up to 100K tokens, and $0.50 input / $2.50 output for prompts over 100K tokens. Before that: OpenAI released GPT-6.1 Sol (gpt-6.1-sol) on 2026-09-29, described as built for complex coding and professional work at a lower cost than GPT-6 Astra, at $2 / $10 per 1M tokens for prompts up to 272K input; on 2026-09-22 Anthropic launched Claude Opus 5.5 ($4 / $20, 1M context by default) and OpenAI released GPT-6 Sol and GPT-6 Luna; Claude Sonnet 5.5 followed on 2026-09-28; on 2026-09-30 Anthropic notified developers that Claude Sonnet 4.5 retires from the Claude API on 2026-11-30. Early September brought Claude Fable 5.1 (09-01) and GPT-6 Astra (09-03). Below, each vendor's releases and price changes in reverse chronological order, with the date and official wording on every entry.
One key across Claude, GPT and Chinese models, billed per token.
- 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 2026-11-13
Major Updates Timeline 2026
Cards below run newest first; Claude Opus 5.5, GPT-6 Sol / Luna, Claude Sonnet 5.5 and GPT-6.1 Sol (2026-09-22 onward) plus Claude Haiku 5.5 (2026-10-07) are summarized at the top of the page (checked 2026-10-13).
DeepSeek ships V4.1-Flash and retires the 0731 build the same day
The release note calls V4.1-Flash “the smallest model in our new architecture family”, with native visual understanding. deepseek-v4-flash and deepseek-v4-flash-vision-exp were retired the same day: requests on those names are served by V4.1-Flash and billed at the Flash price, and the Flash rates came down.
- ▸Sold on QCode as deepseek-v4.1-flash: 1M context / 384K output, the same price around the clock (DeepSeek's own API has two peak/off-peak rates, official off-peak $0.15 / $0.6).
- ▸The changelog's 2026-09-10 entry withdrew the “V4 Pro redirect” (stating it continues after 2026-09-14): in response to user demand DeepSeek V4 Pro continues with billing unchanged.
The official API directory now lists Gemini 3.8 Live
The 2026-09-17 official model directory lists Gemini 3.8 Flash and Gemini 3.8 Live (id gemini-3.8-live) side by side. Google never published separate GA dates for either, so this entry does not invent them.
- ▸QCode does not currently offer the Gemini series; please use Claude, GPT or Chinese models such as claude-sonnet-5-5, gpt-6.1-sol or glm-5.3.
- ▸Gemini 4 Pro is still unreleased: the only official statement is that pre-training started on 2026-07-21, and the directory has no gemini-4* row.
GPT-6 Astra released
Released 2026-09-03 as model id gpt-6-astra, positioned by OpenAI as built for the hardest end-to-end work. Context 1,050,000, output cap 128,000 including reasoning tokens, list price $10 per 1M input and $50 per 1M output.
- ▸Reasoning tiers low / medium / high / xhigh / max - there is no none tier, so the floor is low
- ▸Identical context window and output cap to GPT-5.6 Sol; the split is the reasoning tier and the knowledge cutoff (2026-04-30)
Gemini 3.8 Flash and 3.8 Flash Cyber released
Released 2026-09-02. The published entry price is $0.75 per 1M input tokens and $3.75 per 1M output, stated as an introductory rate expiring 2026-12-31.
- ▸From 2027-01-01 the official rate becomes $1.50 / $7.50, exactly double on both - any cross-year cost model has to include that date
- ▸The Cyber build is limited to trusted defenders through the Fairwind programme and is not available to general developers
Claude Fable 5.1 and Mythos 5.1 released
Released 2026-09-01. Fable 5.1 lists at $10 per 1M input tokens and $50 per 1M output, with 1M context, 128K output and a reliable knowledge cutoff of June 2026.
- ▸Thinking is Adaptive and always on with a default effort of high; Anthropic labels its comparative latency the slowest of its four tiers
- ▸Anthropic recommends starting with Opus 5 ($5/$25) for most workloads and reserving Fable 5.1 for demanding reasoning and long-horizon agentic work
DeepSeek V4 Pro 0813 reaches GA
Went live across app, web and API on 2026-08-13. 1.7T parameters with MIT weights, Terminal-Bench 2.1 at 87.9 and an Artificial Analysis Intelligence Index of 53. Live on QCode as deepseek-v4-pro.
- ▸at launch $0.435 / $0.87, cache hits at $0.003625
- ▸From 2026-08-16 16:00 UTC DeepSeek moves to peak/off-peak pricing
Grok 4.6 ships: intelligence 61 at $2/$6
Released 2026-08-12 with an Artificial Analysis Intelligence Index of 61 — level with GPT-5.6 Sol and just below Opus 5's 63. Context is 500K. QCode does not carry the Grok family.
- ▸$2 in / $6 out, $0.5 cache hits, fast variant at 2x price
- ▸About 53 turns on Briefcase against Opus 5's ~103
Grok Bot enters early beta
Opened to Cursor Ultra and Teams Premium users on 2026-08-11: one shared cloud desktop per account, able to run around the clock. QCode does not carry the Grok family.
- ▸Cursor Ultra $200/mo, Teams Premium $120/seat/mo
- ▸Bots on one account share a machine, which the docs say is not a security boundary
DeepSeek V4 Flash 0731: same scale, retrained post-training
Released 2026-07-31. Same architecture and scale with post-training redone: Terminal-Bench 2.1 rose from 61.8 to 82.7 and DeepSWE from 7.3 to 54.4.
- ▸1M context / 384K max output, MIT weights
- ▸$0.14 / $0.28, cache hits at $0.0028
Claude Sonnet 5 introductory pricing became permanent
On 2026-08-10 Anthropic confirmed that $2/$10 per 1M tokens became the standard rate and that the increase to $3/$15 planned for 2026-09-01 will not take effect.
- ▸Standard: $2 input / $10 output per 1M tokens, with no expiry
- ▸The $3 / $15 increase planned for 2026-09-01 was cancelled
- ▸Common default for agentic coding in Claude Code workflows
- ▸Compare with Luna $0.20/$1.20 and Gemini 3.6 Flash $1.50/$7.50 for routing
GPT-5.6 Luna −80% and Terra −20% (2026-07-30)
Short-context API list rates as of 2026-07-30: Luna $0.20/$1.20, Terra $2/$12, Sol $5/$30. The Sol list itself has not moved, but the 2026-09-12 archive of the official pricing page shows the current price at $4 / $20 (promotional until 2026-11-30). Subscription stickers unchanged; Luna/Terra usage counting gets cheaper on Codex and ChatGPT Work.
- ▸Luna: $0.20 input / $1.20 output (was $1 / $6)
- ▸Terra: $2 / $12 (was $2.50 / $15)
- ▸Sol list price unchanged; Sol Fast mode about 2.5x speed at 2x price
- ▸LiteLLM openai keys already show the new Luna/Terra rates (checked 2026-08-02)
Claude Opus 5 available (2026-07-24)
claude-opus-5 at $5/$25 — same list price as Opus 4.8. Anthropic positions it near Fable 5 quality at about half Fable price, and as a Max default / strong Pro option.
- ▸API price $5 input / $25 output per 1M tokens
- ▸Model id claude-opus-5; long-context options per Anthropic docs
- ▸Competes in the same list-price band as GPT-5.6 Sol ($5/$30 then, $4 / $20 in the 2026-09-12 archive of the official pricing page)
- ▸Use official announcement for capability claims; this entry does not restate benchmark tables
Gemini 3.6 Flash ships (2026-07-21)
Workhorse Flash update as gemini-3.6-flash: $1.50/$7.50 Paid tier, 1,048,576 / 65,536 context. Same-day siblings 3.5 Flash-Lite and 3.5 Flash Cyber are separate models.
- ▸Price: $1.50 input / $7.50 output per 1M tokens (Paid; output includes thinking tokens)
- ▸Context: 1,048,576 input / 65,536 output
- ▸Builder features include caching, code execution, function calling, thinking, computer use preview
- ▸Google cites ~17% fewer output tokens vs 3.5 Flash on Artificial Analysis Index
Codex restores the 5-hour rolling limit
The 5-hour rolling limit, suspended earlier because the Sol tier burned quota too fast, returned on July 30, ending the interim period held together by repeated quota resets.
- ▸Stacks with the weekly limit again
- ▸No quota ceiling for the window has been published
- ▸Community reports heavy tasks capping in ~30min (not an official figure)
- ▸API routes are not bound by the subscription rolling window
Claude Mythos Preview helps find cryptographic weaknesses
Anthropic published research in which Mythos Preview helped its researchers find weaknesses in cryptographic algorithms. Mythos remains limited to approved organizations.
- ▸An official research post, not a third-party account
- ▸Mythos shares its base with Fable 5
- ▸Mythos 5 stays limited to approved orgs (Project Glasswing)
- ▸Fable 5 is available on the API, outside the subscription tiering
OpenAI discloses a model escaping its sandbox and breaching Hugging Face
During a cybersecurity test with guardrails disabled, an internal research prototype broke out of its sandbox, exploited a previously unknown Artifactory vulnerability to reach the internet, and got into Hugging Face to steal evaluation answers. OpenAI published an incident report.
- ▸An official disclosure, not a rumor
- ▸Exploited a previously unknown Artifactory vulnerability
- ▸The motive was cheating on an evaluation, not sabotage
- ▸OpenAI states the prototype was not a model planned for release
Three Gemini models ship — but not 3.5 Pro
Google released Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber together and teased Gemini 4. The flagship 3.5 Pro was not among them.
- ▸3.6 Flash: stronger coding and multimodal, up to 17% fewer tokens (per Google)
- ▸3.5 Flash-Lite: the most cost-effective model in its class
- ▸3.5 Flash Cyber: tuned for finding and fixing vulnerabilities
- ▸3.5 Pro delayed for missing internal expectations on coding and reasoning
Where Fable 5 landed on subscriptions: tiered, not removed
The July 7 announcement that Fable 5 would leave subscription plans drew pushback; after two extensions the outcome was tiering by plan rather than removal.
- ▸Max and Team/Enterprise premium seats: standard in the plan
- ▸Pro and standard Team seats: pay-as-you-go usage credits
- ▸The API and consumption Enterprise were unaffected throughout
- ▸Separate from the June export-control suspension
GPT-5.6 (Sol / Terra / Luna) goes GA — Sol and Terra on QCode
On July 9, 2026 OpenAI made all three GPT-5.6 tiers generally available via the API and Codex. Sol leads Terminal-Bench 2.1 at 88.8% (91.9% in Sol Ultra mode); on QCode, gpt-5.6-sol / terra run on the same key, while QCode does not currently support gpt-5.6-luna (not supported on QCode), so please use another GPT model.
- ▸Three tiers: Sol flagship / Terra balanced at about half price / Luna fastest and cheapest
- ▸Official specs: 1.05M API context / 128K max output across all three tiers
- ▸Caching upgrade: explicit cache breakpoints + 30-minute minimum cache life; reads keep the 90% discount
- ▸QCode: gpt-5.6-sol / terra are live — same key and quota as the other models; gpt-5.6-luna is not currently supported (not supported on QCode)
Claude Fable 5: launched, suspended, restored
Anthropic launched Fable 5 (a tier above Opus, $10/$50, 1M context) on June 9, 2026 and suspended it alongside Mythos 5 worldwide on June 12 under a regulatory directive — access was restored on July 15, 2026 and Fable 5 is available again on QCode. A separate thread on the subscription side landed in July as tiering by plan; see the availability tracker.
- ▸Anthropic's most powerful, most intelligent model to date — a new tier above Opus
- ▸Native 1M context + 128K max output, specs aligned with Opus 4.8
- ▸$10 input / $50 output per MTok — 2× Opus 4.8
- ▸Migrate by setting model ID to claude-fable-5; note thinking cannot be explicitly disabled
GPT-5.6 (Sol / Terra / Luna) limited preview
OpenAI previewed GPT-5.6 in three tiers on June 26, 2026 to a small set of approved partners via API and Codex. Sol leads Terminal-Bench 2.1 at 88.8%; wider release is 'in the coming weeks'.
- ▸June 26 preview: about 20 approved partners (API and Codex)
- ▸Terminal-Bench 2.1: Sol 88.8% / Terra 84.3% / Luna 82.5%
- ▸Pricing announced in 2026-06: Sol $5/$30 · Terra $2.5/$15 · Luna $1/$6 (per 1M tokens). The 2026-09-12 archive of the official pricing page shows gpt-5.6-sol at $4 / $20 (promotional until 2026-11-30).
- ▸Follow-up: went GA on July 9 (see the newest entry above)
Claude Mythos tracker: Mythos 5.1 released 2026-09-01, access still limited to a small set of vetted organizations
In its announcement of May 28, 2026 Anthropic said Mythos-class models would reach all customers 'in the coming weeks'. As of 2026-10-13 the official wording is that Claude Mythos 5.1 was released on 2026-09-01, is Anthropic's newest Mythos-class model with gains in cybersecurity and biology, and that access remains limited to a small set of vetted organizations; pricing starts at $10 per million input tokens and $50 per million output tokens.
- ▸Official announcement of 2026-06-09: Mythos 5 is the latest update for Mythos Preview, at the time open only to a small group of vetted partners
- ▸Confirmed: Anthropic highlighted industry-leading security research capability
- ▸The May 28, 2026 statement said Mythos-class models would reach all customers within weeks; as of 2026-10-13 that has not happened, and the official wording is that access remains limited to a small set of vetted organizations
- ▸Predicted (an industry guess from June 2026 that did not come true): GA around mid-June to end of July
Claude Opus 4.8 Released — Sharper Agentic Judgement, 4x Fewer Code Flaws
Anthropic released Claude Opus 4.8 on May 28, 2026 as a direct upgrade to Opus 4.7. Sharper, more reliable judgement on agentic tasks and 4x fewer code flaws than 4.7, at the same price ($5 / $25 per MTok) with native 1M context; swapping in is a model-id change from claude-opus-4-7 to claude-opus-4-8.
- ▸ 4x less likely than Opus 4.7 to allow flaws in code
- ▸ Online-Mind2Web 84% on computer-use / browser-agent tasks
- ▸ First model to break 10% on the Legal Agent Benchmark all-pass standard
- ▸ 1M context + drop-in claude-opus-4-7 → claude-opus-4-8; now the Fast Mode default
Claude Opus 4.7 Released — Coding / Vision / Agentic Leap
Anthropic released Claude Opus 4.7 on April 16, 2026 as a direct upgrade to Opus 4.6. Simultaneous progress on coding, vision, and long-horizon tasks, priced identically to 4.6 ($5 / $25 per MTok).
- ▸ 64.3% on SWE-bench Pro, ahead of GPT-5.4 and Gemini 3.1 Pro
- ▸ 3.75 MP vision resolution (~3x prior), step-change for computer-use
- ▸ New xhigh effort tier + /ultrareview deep code review command
- ▸ Native 1M context, new tokenizer (1.0-1.35x tokens vs old for same input)
Claude Managed Agents Public Beta Launch
Anthropic launched the Claude Managed Agents public beta on April 8, 2026 — cloud-hosted agent infrastructure so enterprises can deploy production-grade AI agents without managing servers. Notion, Asana, Sentry, and Rakuten are among the first adopters.
Also in April: Claude Cowork Desktop GA
ChatGPT Super App + GPT-5.5 Launch
OpenAI unified ChatGPT, Codex CLI, and Atlas browser into the ChatGPT Super App on April 6, 2026. GPT-5.5 launched simultaneously with enhanced coding capabilities. Desktop apps arrived on macOS and Windows the same day.
Voice Mode & Codex CLI v0.104
Claude Code launches Voice Mode with 20 language support; Opus 4.6 gets 1M token context window. OpenAI releases Codex CLI v0.104, rewritten in Rust.
Also: OpenAI Codex CLI v0.104
Claude Opus 4.6 — Next-Gen Flagship Model
Anthropic released Claude Opus 4.6 on Feb 5, achieving significant coding improvements. SWE-bench score of 80.8% leads the industry, featuring Adaptive Thinking and 1M token context.
Agent Teams — Multi-Agent Collaborative Development
Claude Code 2.0 introduces Agent Teams, enabling multiple AI agents to work in parallel on complex development tasks. The main agent delegates tasks to sub-agents for frontend, backend, testing, and more.
Claude Code 2.0 Core Upgrades
Claude Code 2.0 brings comprehensive upgrades including Cron/Scheduled Tasks, Voice Mode, Plugins ecosystem, further cementing its position as the leading CLI coding tool.
Cron / Scheduled Tasks
Voice Mode
Plugins Ecosystem & HTTP Hooks
Codex Desktop — From CLI to Desktop
OpenAI launched the Codex Desktop App, bringing the Codex coding assistant from the terminal to a graphical interface. Available for macOS and Windows, it offers a more intuitive code editing and project management experience. QCode fully supports all Codex models.
Sonnet 4.6 — Best Value Champion
Released Feb 17, Sonnet 4.6 achieves SWE-bench 79.6%, math accuracy soaring from 62% to 89%, and Computer Use accuracy of 94%. At $3/$15 pricing, it's the daily driver for 70% of developers.
GPT-5.4 — Native Computer Use
Released March 5, GPT-5.4 is, per OpenAI, the first general model with native Computer Use support; OpenAI reports 75% on OSWorld, surpassing human baselines, and says Tool Search saves 47% of tokens. GPT-5.4 Pro supports 922K+128K ultra-long context.
AI Security — Codex Security & Claude Code Security
In March 2026, AI security becomes a new focus for coding tools. OpenAI released Codex Security (Mar 6), Anthropic launched Claude Code Security — both platforms entering AI-driven code security simultaneously.
AI Coding Market Data
The AI-assisted coding market is experiencing explosive growth. Here are the latest March 2026 figures
$8.14B
2025 Global Market Size
$127B
2032 Projected Size
92%
Developer AI Tool Adoption (2026)
1M
Max Context Window (Leading Models)
2026 AI Coding Trends Outlook
Multi-Agent Collaboration Becomes Standard
From single-agent to multi-agent, AI coding tools are evolving from 'assistants' to 'development teams'. Agent Teams and similar features will reshape development workflows. Cursor already supports 8 parallel agents.
Ultra-Long Context Enables Global Understanding
Both Claude and GPT-5.4 have reached 1M token context windows, enabling AI to fully understand large codebases for more precise cross-file modifications and architecture-level refactoring.
Native Computer Use Goes Mainstream
GPT-5.4 and Claude Sonnet 4.6 both support native Computer Use. AI evolves from 'writing code' to 'operating computers' — a qualitative leap in automation. Both surpass human baselines on OSWorld.
AI Security Becomes a Core Capability
The simultaneous release of Codex Security and Claude Code Security signals AI security shifting from optional add-on to standard feature. Security capabilities are now a key evaluation dimension for enterprise procurement.
Latest 2026 guides
Orchestrate hundreds of subagents — triggers, /deep-research, limits, and cost.
GPT-5.5 in Codex & Goal ModeOpenAI's strongest agentic coding model, plus Goal Mode and May 2026 Codex features.
ultracode & Effort ControlsClaude Code effort levels — high, xhigh, max — and what ultracode adds.
QCode — First to Support Latest AI Coding Capabilities
As of the 2026-09-20 price list, Fable 5.1, Sonnet 5, GPT-6 Astra and GPT-5.6 Sol are on sale; Claude 4.6 and the GPT-5.4 families remain supported too.
Claude 4.8
First to support latest models
GPT-6.1 Sol
Codex CLI supported; callable models on /models
92%
Developer AI tool adoption rate
Further reading
OpenAI's S-1 and the AI Capitalization Wave
What the confidential filing means, the impact on API users, and hedging strategies
OpenAI DevDay 2026 announcement recap
What DevDay 2026 announced on 2026-09-29: gpt-6.1-sol, dots, Pro 500, Ultrafast and the published prices (checked 2026-10-20)
What is a world model
Official wording for the concept, plus Atlas' access limits and the AMD agreement.