Core References

AI Model Reference Guide

Compare current AI model families, understand reasoning and sampling controls, and choose models for research reasoning, writing, and analysis tasks.

The available models change fast; the task-fit question stays constant.

Overview: current, but not permanent

Warning

AI model names, aliases, pricing, rate limits, and control parameters change quickly. The families, controls, access qualifications, and retirement notices below were reviewed against public documentation on October 19, 2026. Documentation-confirmed; account API access not tested. A provider's current model is not necessarily available in your application's menu. Before building a workflow around a specific model ID, check the provider's current model list and your client's supported route.

The practical question is not "which model is best?" It is which model, access path, and control surface fit the task:

Reasoning depth: Can it handle theoretical tension and multi-step analysis?
Context capacity: Can it read the relevant corpus without destructive chunking?
Writing behavior: Does it preserve academic nuance and voice?
Cost and latency: Can you afford to iterate?
Control surface: Does it want reasoning effort, thinking level, prompt constraints, or sampling parameters?

Choose a model for the task

Choose for the work you need done. These suggestions reflect editorial judgment and the author’s experience, not measured rankings.

Kimi is a regular collaborator in the author's reasoning, writing, coding, and sustained project work. Its different perspectives matter from the first approach onward. The roles below can change with the task.

Good starting points:

Claude Opus 5.5 or Claude Sonnet 5.5 for theory, critique, nuanced writing, and focused implementation.
GPT-6 Astra or Kimi K3 when the same model may also need to coordinate later.
Gemini 3.8 Flash for large-context synthesis. Gemini 4 Argon is a newer paid candidate to compare where your client supports it, not an author-tested replacement.
GLM-5.3 for long-context coding-agent reasoning and Chinese-English synthesis.

Tasks: theory building, critical analysis, synthesis, methodology design, and one scoped coding task.

Keep a model that works for your task while the provider still supports it. Compare a replacement when quality, cost, access, or behavior gives you a reason to change.

See the AI Model Reference Guide for the full comparison.

Test these suggestions against your own material. Check support and retirement dates before keeping an older model.

Current model families

OpenAI GPT Family

Current flagship provider-native ID: gpt-6-astra, alongside gpt-6.1-sol and gpt-6-luna. The GPT-5.6 family remains available as gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna; the gpt-5.6 alias routes to Sol, not Astra.
Context: roughly 1M on Astra, Sol, Terra, and Luna. Check the model catalogue (opens in a new tab) for exact limits.
Control surface: reasoning.effort where supported. Astra and GPT-6.1 Sol support low through max, but not none; Sol defaults to medium and uses Responses for tool calling. Check the exact model's controls before reusing an older configuration.
Suggested collaboration: Astra or GPT-6.1 Sol for research and implementation across a project; current Luna for volume work. Keep older Terra where it already works well. Compare Sol with Astra on the same task, as OpenAI’s model selection guide (opens in a new tab) recommends.
Source: OpenAI latest-model guide (opens in a new tab)

Anthropic Claude Family

Current provider-native IDs: claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5, and claude-haiku-5-5. Haiku 5.5 is the current faster, lower-cost option. Haiku 4.5 remains active as a legacy option; an older ID is not necessarily retired.
Context: 1M on Fable 5.1, Opus 5.5, and Sonnet 5.5.
Control surface: use effort and thinking controls where documented. Fable 5.1 and Opus 5.5 always use adaptive thinking; Sonnet 5.5 supports it; keep sampling defaults unless the current model page says otherwise.
Suggested collaboration: Fable 5.1 to coordinate agents; Opus 5.5 or Sonnet 5.5 for a defined analysis, writing, or implementation task. Sonnet 5.5 replaces older Sonnets in these recommendations.
Haiku 5.5 has a 1M context window, up to 128K output, and adaptive thinking on by default with medium effort. Its pricing (opens in a new tab) changes above 100K input tokens. See the Haiku model guide (opens in a new tab) before migrating manual-thinking settings from 4.5.
Retirement notice: Sonnet 4.5 is deprecated and scheduled to retire November 30, 2026. An application's saved model selection does not extend provider support (opens in a new tab).
Source: Claude model docs (opens in a new tab)

Kimi Family (Moonshot AI)

Current flagship provider-native ID: kimi-k3.1-pro. Kimi K3 remains supported as kimi-k3; a newer release does not invalidate the author's earlier experience.
Context: 1M on Kimi K3.1 Pro and K3; 256K on K2.7 Code and K2.6.
Access and lifecycle: K2-series direct-API access is unavailable to newly registered users; users registered before November 1 can continue for now. Moonshot describes deprecation as expected in late December 2026, not a guaranteed retirement date. Gateway availability is a separate question.
Direct API access: K3 requires a successful top-up of at least $1. Account tier affects rate limits. This is separate from Kimi Code subscription access.
Control surface: Kimi K3 always thinks and accepts reasoning_effort values low, high, and max; the default is max. Omit sampling parameters.
Suggested collaboration: Kimi K3 as either a direct worker or a multi-model coordinator, based on the author's experience. Compare K3.1 Pro on your own task; it is a current provider offering, not a new comparative result here. K2.7 Code is a legacy option for existing accounts, not a new-account starting point.
Author's experience (October 12, 2026): Kimi is a regular collaborator in reasoning, writing, coding, and sustained project work. Its different perspectives are the main reason to engage with it across these tasks. See a workflow that starts with Kimi.
Sources: Kimi model list and lifecycle (opens in a new tab), Kimi K3 quickstart (opens in a new tab). Documentation-confirmed; account API access not tested.

Google Gemini Family

Current choices: gemini-4-argon is Google's current flagship. gemini-3.8-flash remains a stable Flash choice; gemini-3.6-flash remains listed; gemini-3.5-flash-lite serves volume work; gemini-3.1-pro-preview remains in preview. Google has deprecated gemini-3.7-flash and routes its requests to 3.8 (opens in a new tab). An accepted ID does not prove that the original model still runs.
Argon context: input, thinking, and output share 1,048,576 tokens. Synchronous output is capped at 65,536; longer output needs the documented long-decode workflow. See the model specifications (opens in a new tab).
Control surface: omit temperature, top_p, and top_k on Argon; use thinking_level (low, medium, or high, default medium). Explicit context caching is unsupported. Current Gemini 3.x releases also deprecate sampling controls; Gemini 3.8 Flash does not support minimal.
Argon access: paid Gemini API and AI Studio, with no Free tier. Introductory pricing is $2 per million input tokens and $10 per million output tokens through November 30, 2026, then $4 and $20. Check current pricing (opens in a new tab) and client support before switching; a large output budget can be costly.
Best for: large-corpus synthesis, volume processing, multimodal tasks, Google platform workflows.
Author's assessment (October 12, 2026): The author reports a much better experience working with Gemini 3.8 Flash. This supports keeping it as a starting choice. The assessment does not establish which tasks improved, and is not a comparison with Argon.
Source: Gemini latest-model guide (opens in a new tab)

DeepSeek Family

Current provider-native anchor: deepseek-v4-pro and deepseek-flash (DeepSeek V4.1 Flash).
Migration note: deepseek-v4-flash is a legacy compatibility string; DeepSeek's current documentation says the corresponding model is retired and requests are served by V4.1 Flash. Use deepseek-flash for new direct-provider configuration.
Control surface: current V4 models use thinking mode plus reasoning_effort; the default thinking mode is enabled.
Best for: high-volume exploration, budget-sensitive screening, reasoning checks where cost matters.
Source: DeepSeek model list (opens in a new tab) and current model/pricing docs (opens in a new tab)

GLM Family (Z.ai / Zhipu AI)

Current provider-native ID: glm-5.3 (GLM-5.3).
Context/output: 1M context and up to 128K output.
Account/endpoint limit: Z.ai currently restricts accounts with any previous GLM Coding Plan subscription, including an expired one, to the Chat-Completions-compatible model API route. Check the documented endpoint for your account rather than assuming the Responses or Anthropic-compatible route is interchangeable.
Control surface: GLM-5.3 always uses thinking; reasoning_effort accepts low, high, or max, defaulting to max.
Best for: long-context coding-adjacent reasoning, multilingual Chinese-English synthesis, frontier-adjacent open-weight experiments.
Source: GLM-5.3 docs (opens in a new tab)

Qwen Family (Alibaba)

Current provider-native IDs: qwen3.8-max, qwen3.7-plus, and qwen3.8-flash. Flash is the lower-cost option.
Context: current Max, Plus, and Flash entries support up to 1M in the documented deployment modes.
Lineup note: Max is the difficult-task model; Plus balances capability, speed, and cost; Flash is the volume option.
Caveat: model names and availability differ by international, global, US, EU, Hong Kong, and Chinese mainland deployment modes.
Best for: multilingual work, open-source comparisons, region-specific provider setups.
Source: Alibaba Model Studio model list (opens in a new tab)

xAI Grok Family

Current provider-native ID: grok-4.7 (Grok 4.7) for general and coding work.
Context: Grok 4.7 lists a 500K-token context window.
Control surface: configurable reasoning; check the current model's supported effort levels before carrying over a previous configuration.
Best for: agentic tool calling, coding, broad general tasks, xAI platform experiments.
Author's experience (October 12, 2026): Grok 4.6 remains the more familiar choice. The author works with 4.7 less often and has not judged it better or worse.
Source: xAI model docs (opens in a new tab)

Local / Open-Weight Models

Current anchor: depends on your runtime, hardware, and privacy constraints.
Best for: sensitive data, reproducibility, offline work, customization.
Caveat: local quality depends more on deployment details than provider marketing pages.

Access paths

Provider-native IDs, gateway-prefixed IDs, client aliases, and subscription menus are different surfaces. A model list or configured alias is not proof of account entitlement. GPT-5.5 retired from Codex with ChatGPT sign-in (opens in a new tab) on October 14, 2026; its OpenAI API availability and API-key-authenticated Codex sessions were not affected.

Access path	Best for	Notes
Provider web app	Fast exploration	Good for trying a model before wiring API access.
Google AI Studio	Gemini API keys and selected free-tier experiments	Argon is paid. Use for direct Gemini model access, not Antigravity CLI auth.
Cherry Studio	GUI model comparison and knowledge bases	Best first interface for many researchers.
OpenCode	CLI-side provider comparison	Useful when model choice is part of the method.
Vox MCP	Multi-model access inside an MCP client	Pure passthrough; good for challenge and triangulation.
Claude Code / Antigravity CLI	Agentic project work	Default product sign-in, with documented API/provider routes also available.

Configuration: defaults first

Sampling is no longer the default lever

The old advice was to tune temperature for every model. That is now misleading. Many current reasoning models either reject sampling changes, ignore them, or explicitly recommend defaults.

Default rule: omit temperature, top_p, and top_k unless the provider currently documents that the model supports and benefits from changing them.

Provider-specific notes:

Provider/model	Recommendation
Claude 5 family	Keep sampling defaults; use `effort` and the model's documented thinking controls.
Gemini Flash / Flash-Lite	Use the exact model's documented thinking controls; omit deprecated sampling parameters.
Kimi K3 / K2.7 / K2.6	Omit sampling parameters; use `reasoning_effort` on K3.
GLM-5.3	Always-on thinking; use `reasoning_effort` (`low`, `high`, or `max`).
DeepSeek V4	Use thinking mode and `reasoning_effort`; keep other controls at provider defaults.
xAI Grok 4.7	Prefer documented reasoning controls; watch incompatible parameters on reasoning models.

If a compatible model requires a temperature and the provider offers no task-specific guidance, start at 1.0. Go lower only for mechanical extraction or classification where repeatability matters.

Prefer reasoning controls when available

Control	Providers/examples	Use for
`reasoning.effort` / `reasoning_effort` / `effort`	OpenAI, Claude, Kimi K3, xAI, GLM, DeepSeek	More or less deliberate reasoning.
`thinking_level`	Current Gemini models	Gemini reasoning depth without raw sampling changes.
thinking mode	Kimi K2.6, DeepSeek V4	Choosing thinking behavior where supported.
Sequential Thinking MCP	Any MCP-capable client	Tool-visible, revisable reasoning steps.

Use high reasoning effort for theoretical integration, methods design, and hard debugging. Use low/default effort for routine extraction, tagging, and quick checks.

Task type	Preferred control	Suggested starting point
Theory synthesis	Reasoning effort / thinking level	Medium or high; keep sampling defaults.
Framework building	Reasoning effort plus explicit criteria	High; ask for alternatives and tensions.
Literature synthesis	Context capacity plus structured prompt	Medium/high thinking; defaults for sampling.
Draft writing	Prompt constraints and examples	Defaults first; ask for voice and structure directly.
Citation extraction	Deterministic prompt rules	Low reasoning or provider-specific low temp if supported.
Systematic coding	Rubric and validation pass	Low/medium reasoning; prioritize consistency checks.

Working across models

Use cheap or free access paths to learn model behavior against your own corpus.

Start with Google AI Studio, Cherry Studio, OpenCode, or OpenRouter.
Run the same prompt across at least three model families.
Record failure modes, not just best answers.
Promote a model only after it handles your real material.

Work shape	Good starting models	Why
One bounded theory, critique, or writing task	Claude Opus 5.5 or Claude Sonnet 5.5	Start here for a defined task.
Carry out an agreed project plan	GPT-6 Astra or GPT-6.1 Sol	The author finds they follow through more often once criteria and boundaries are explicit.
Develop another approach from the outset	Kimi K3	The author values Kimi's different perspectives across reasoning, writing, coding, and sustained work.
Multi-model coordination	Claude Fable 5.1, GPT-6 Astra, or Kimi K3	Starting points when the session must direct several models.
Alternating between coordination and direct work	GPT-6 Astra or Kimi K3	Both support coordination and substantive direct work.
Large literature sets	Start with Gemini 3.8 Flash; compare Gemini 4 Argon or GLM-5.3 where available	Argon is a new candidate, not an author-tested upgrade. Check synthesis quality and cost on your own papers.
Lower-cost screening	DeepSeek Flash, Gemini Flash or Lite, Qwen Plus or Flash	Compare errors as well as cost.

Free and low-cost options

Google AI Studio offers free access to selected Gemini models, not the entire catalogue; Argon has no Free tier. Check Gemini pricing (opens in a new tab) and rate limits (opens in a new tab) before relying on a quota for teaching or batch work.

Other cost-management paths:

Use OpenCode or Cherry Studio to compare providers before committing to a long run.
Work with DeepSeek or Qwen on high-volume screening when quality is sufficient.
Work with local models when privacy matters more than frontier quality.
Bring higher-cost models into the stages where their contributions justify the expense.

Next step

The reference above tells you what is available. The companion page tells you how to test those choices against your own materials: AI Model Discovery Protocol.

cite this page

Lin, X. (2026). AI Model Reference Guide. Research Memex. https://research-memex.org/docs/implementation/core-references/ai-model-reference-guide

@misc{docs-implementation-core-references-ai-model-reference-guide-2026,
  author = {Xule Lin},
  title = {AI Model Reference Guide},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/implementation/core-references/ai-model-reference-guide}},
  note = {ORCID: 0000-0001-7885-4194}
}

one renderingthe source remains