OpenAI's most capable model — 1.05M context, 128K output
Announced September 3, 2026, and the top of the GPT-6 family. $10/$50 per million tokens puts it in the same price band as Claude Fable 5.1, so treat it the same way: a model for the hardest reasoning and long-horizon agent work, not a default. Prompts over 272K input tokens bill at a higher long-context rate ($20/$75). Most teams should start on GPT-6.1 Sol and only move up when evals show Sol is the bottleneck.
The OpenAI default — near-Astra quality at a fifth of the price
Shipped September 29, 2026, one week after GPT-6 Sol, and replaced it at the same $2/$10 list price. OpenAI positions it as near-Astra on agentic coding, computer use, and professional work, with the same 1.05M-token context and 128K output. This is the sane OpenAI default for agentic and coding workloads — the role GPT-5.6 Terra held over the summer.
OpenAI's efficient tier — $0.10/$0.50 with the full 1.05M context
Launched September 22, 2026. The default now for classification, extraction, summarisation, routing, and simple chat: $0.10 input / $0.50 output per million tokens, with the same context window as the flagship. It undercuts GPT-4o mini on price while sitting two model generations ahead, which removes the last cost argument for staying on the 4o or GPT-5 mini tiers.
Previous OpenAI family — Sol/Terra/Luna tiers, superseded by GPT-6
GA in July 2026 and the OpenAI default until the GPT-6 family landed in September. Still served (Sol $4/$20 on promotional pricing, Terra $2/$12, Luna $0.20/$1.20), and still the newest OpenAI family on Azure, where it is the named replacement for retiring GPT-4o and o-series deployments. On the OpenAI API, GPT-6.1 Sol and GPT-6 Luna are both better and cheaper — migrate there. On Azure, this is what you deploy today.
OpenAI's 2025 flagship — two families back now
Still served at $1.25/$10, but two families behind: GPT-6.1 Sol beats it on quality for $2/$10 with a context window more than twice the size. No reason to build new on plain GPT-5 — migrate to GPT-6.1 Sol.
OpenAI's 2025 cost-efficient model — superseded by GPT-6 Luna
Was the default for classification, extraction, and simple chat through early 2026. At $0.25/$2.00 it now costs more than twice what GPT-6 Luna does ($0.10/$0.50) for an older model with a smaller context window. Fine to leave in a pipeline that meets its bar; pick Luna for anything new.
OpenAI 2024 multimodal model — legacy, with retirement dates on Azure
Still served on the OpenAI API and still running a lot of production RAG, but it is now a legacy model with a clock on it: Azure lists every GPT-4o version as deprecated or legacy, with retirements between December 2026 and April 2027. A 128K context and $2.50/$10 pricing also compare badly with GPT-6.1 Sol at $2/$10 and 1.05M. If you are still on it, plan the migration and re-run your evals now rather than at the deadline.
OpenAI's 2024 cost-efficient model — superseded by GPT-6 Luna
GPT-6 Luna is cheaper ($0.10/$0.50 against $0.15/$0.60), far more capable, and has roughly eight times the context window, so there is no remaining reason to pick this for a new build. Azure lists it as deprecated with an April 2027 retirement. Migrate when you next touch the pipeline.
OpenAI's first open-weight model — 120B params, fully downloadable
OpenAI's 2025 open-weight release. Competitive with Llama 4 and Mistral Large 3. Fine-tunable and self-hostable, which matters more now that OpenAI is winding down its hosted fine-tuning platform. A good fit for teams that want an OpenAI-lineage model with full control.
Anthropic's most capable widely released model — for the hardest reasoning and longest agent runs
The top of the Anthropic lineup, above Opus. Fable 5.1 succeeds Fable 5 at the same $10/$50 per million tokens, with a 1M context and 128K output. Thinking is always on and you steer depth with an effort setting rather than a token budget; forced tool choice is rejected, so tool use has to be steered from the prompt. Reach for it on architecture review and long-horizon agent work, and expect single requests on hard tasks to run for minutes. For most work, Opus 5.5 is the better cost-to-quality pick.
The current Opus — agentic coding and computer use, now cheaper than Opus 5
Released September 22, 2026, replacing Opus 5 (July 2026) with a price cut: $4/$20 per million tokens against $5/$25, same 1M context and 128K output. This is the default Anthropic model for demanding work, and the better cost-to-quality pick over Fable 5.1 when the task is execution — agentic coding, computer use — rather than open-ended reasoning. Migration notes if you are coming from Opus 5: thinking can no longer be disabled, effort now defaults to medium, and forced tool choice returns an error.
Anthropic's production workhorse — speed and capability for everyday coding and agents
The one most teams should actually be running in production. Released September 28, 2026, replacing Sonnet 5 at the same $2/$10 per million tokens with a 1M context and 128K output. MCP tool ecosystem is mature. First pick for agentic workflows, code generation, RAG, and day-to-day engineering assistance. Coming from Sonnet 5: disabling thinking and forcing a tool choice both return errors now, and effort levels were recalibrated, so re-run your effort sweep.
Anthropic's fast tier — $0.10/$0.50 with a 1M context
Released October 7, 2026, and the biggest pricing move in the Anthropic lineup this year: $0.10 input / $0.50 output per million tokens for prompts up to 100K tokens ($0.50/$2.50 beyond), a tenth of what Haiku 4.5 cost, with the context window going from 200K to 1M. Adaptive thinking is on by default. The obvious pick for classification, extraction, routing, LLM-as-judge, and sub-agent work. Haiku 4.5 is still served at $1/$5 but there is no reason to start anything new on it.
Anthropic's 2025 workhorse — superseded by Sonnet 5.5
Several releases back from Sonnet 5.5 now, at a higher price ($3/$15) and a fifth of the context window. Only worth keeping where existing evals are tuned against it and nobody has had time to re-run them. New projects should start on Sonnet 5.5.
Anthropic's 2024 workhorse — legacy
A 2024 model, well behind Sonnet 5.5 on every axis that matters, including price. There is no reason to start a new project here — treat it as a legacy integration you have not migrated yet.
Google's most intelligent Flash model — for coding and agents
Released September 2, 2026, three weeks after 3.7 Flash, which it replaced. The sharpest thing Google ships right now — including against its own Pro tier, which has been frozen at 3.1 since February and is still served under a preview id. $0.75/$3.75 per million tokens with a 1M context, but that is an introductory price: it doubles to $1.50/$7.50 on January 1, 2027, so budget for the higher number. If you are choosing a Gemini model today for coding or agentic work, start here, not with "Pro."
Google's Pro-tier model — capable, but its own Flash line has passed it
Still the current "Pro" release and still a preview id (gemini-3.1-pro-preview): Google has not updated this tier since February 2026 while Flash raced to 3.8 — and 3.5 Flash already outperformed 3.1 Pro on agentic and coding benchmarks per Google's own comparisons. $2/$12 per million tokens up to 200K, $4/$18 beyond. Native 2M context and deep multimodal understanding still make it the right call for very long documents. For coding and agents specifically, pick Gemini 3.8 Flash instead.
Google's previous Pro-tier model — superseded by Gemini 3.1 Pro
A full major version behind the current Pro tier, and Google now limits the whole 2.5 family to accounts that already used it — new projects cannot start here at all. Migrate to 3.1 Pro, or to 3.8 Flash if the workload is coding or agentic.
Google's 2024 speed-optimised model — shut down on the Gemini API
Gone: Google has shut down Gemini 2.0 Flash and 2.0 Flash-Lite on the Gemini API. Anything still pointing at it is failing. Move to Gemini 3.8 Flash, or 3.5 Flash-Lite ($0.30/$2.50) if cost is the constraint.
Meta's new proprietary flagship — closed-weight, agentic-first
Meta's Superintelligence Labs shipped this as a genuinely new direction, not a Llama update — closed-weight, built explicitly for multi-agent orchestration (as either the planning agent or a delegated subagent), with a 1M-token context window and text/image/video/audio/PDF input. The 1.2 update (Aug 5, 2026) added a dedicated coding focus plus Muse Code, a terminal agent. The real tradeoff: you lose everything that made Llama useful for self-hosting — no local deployment, no fine-tuning, no community variants. Pick this for Meta's agentic capability specifically; pick Llama if open weights still matter to you.
Meta's flagship open model — 400B MoE with 128 experts
Massive leap for open-source. 400B MoE architecture with 128 experts runs efficiently on 8× H100. Matches frontier closed models on most benchmarks. Best open model for enterprise self-hosting — and now the more relevant Meta model for that use case, since Meta's newer Muse Spark line is closed-weight.
Meta's proven open model — battle-tested at 70B params
Battle-tested in thousands of production deployments. Runs on a single A100 80GB. Excellent for self-hosted RAG, fine-tuning, and cost-sensitive pipelines. Huge ecosystem of fine-tunes.
1.6T MoE, 1M context, MIT-licensed — DeepSeek's new flagship
Released April 24, 2026, unifying what used to be separate R1 (reasoning) and V3 (general) models into one, with three selectable reasoning modes (Non-Think, Think High, Think Max). 1.6T total / 49B active parameters, a genuinely usable 1M-token context at roughly 27% of the FLOPs and 10% of the KV-cache footprint of the previous generation, MIT-licensed. At $0.435/$0.87 per million tokens it is dramatically cheaper than any closed frontier model at a comparable level — still the best self-hostable option for reasoning-heavy work.
The efficient V4 variant — 284B MoE, same 1M context, MIT license
Same architecture and 1M context as V4 Pro at a fifth of the size (284B total / 13B active) and roughly a fifth of the price ($0.14/$0.28 per million tokens). Give up some ceiling on the hardest reasoning tasks in exchange for materially cheaper, faster inference — the right default for high-volume pipelines that do not need V4 Pro's full capability.
Open-weight chain-of-thought reasoning model — superseded by DeepSeek V4
Made frontier reasoning accessible to everyone when it shipped, and still works — but DeepSeek V4 folded R1's reasoning specialism and V3's general capability into a single model with selectable thinking modes, at a fraction of the inference cost. Migrate to DeepSeek V4 Pro; there is no longer a reason to run R1 and V3 as separate models.
685B MoE model trained for $5.5M — superseded by DeepSeek V4
Proved you can train frontier models affordably, and the efficiency story continues in V4, which replaces it directly with a larger context window, lower inference cost per the DeepSeek team's own figures, and unified reasoning modes. No reason to start new work on V3.
Alibaba's 2.4T multimodal flagship — text, image, and video in one model
Launched August 3, 2026 (2.4T total / 95B active MoE, ~1M-token context, native text/image/video input) at $2/$6 per million tokens. A real step up from the Qwen 3.5 line, not just a version bump — this is now Alibaba's serious multimodal-reasoning play, not just its best multilingual model. Best choice in this catalog for CJK-heavy or multimodal workloads that also need frontier-adjacent reasoning.
Alibaba's previous multilingual model — superseded by Qwen 3.8 Max
Still fine for lightweight multilingual work at 9B–72B sizes with permissive licensing, but Qwen 3.8 Max is a generational jump in both scale and multimodality. Pick 3.8 Max for new projects unless you specifically need Qwen 3.5's smaller, self-hostable sizes.
Google's open model family — 26B and 31B instruction-tuned
Best small-to-medium open model from Google. 26B A4B variant uses mixture of experts for efficiency. Strong at instruction following and reasoning. Good JAX and Keras ecosystem.
675B open-weight MoE, Apache 2.0 — the strongest Mistral model you can download today
675B total / 41B active MoE, 262K context, fully Apache 2.0 — download it, fine-tune it, run it on your own hardware, no license negotiation. At $0.50/$1.50 per million tokens via Mistral's own API it is also cheap if you would rather not self-host. Mistral announced Large 4 on October 6, 2026, but it is API-only for now with weights promised later — until they land, Large 3 is still the Mistral model to self-host for teams that need frontier-adjacent quality and real data-residency control.
Previous Mistral flagship — superseded by Mistral Large 3
Was the best EU-hosted option with data-residency guarantees; Mistral Large 3 replaces it directly and adds a full open-weight Apache 2.0 release on top, so the "EU option" argument no longer requires giving up self-hosting. Migrate new work to Large 3.
Microsoft's 14B model punching above its weight on reasoning
Remarkable reasoning for its size — beats many 70B models on math and logic benchmarks. Runs on consumer GPUs. Ideal for edge deployment and latency-sensitive applications.
xAI's current flagship — same price and context as 4.6
Now the top of the xAI lineup, following Grok 4.6 (August 12, 2026) at an unchanged 500K context and $2/$6 short-context rate. xAI calls it the most capable model it has built; 4.6 and 4.5 are still served at the same price, so there is no cost reason to stay on them. Watch the long-context pricing: once a request reaches 200K prompt tokens, every token in it bills at double rate ($4/$12), not just the overage.
xAI's 2025 model — superseded by Grok 4.7
Three releases behind now (4.5, 4.6, then 4.7). There is no remaining reason to build new on Grok 4 specifically — go straight to 4.7.
Zhipu AI's 754B frontier model from China's leading AI lab
One of the largest dense models available. Strong Chinese language capabilities and general reasoning. Open weights on Hugging Face. Interesting for multilingual and research applications.
Cohere's enterprise model optimised for RAG and tool use
Purpose-built for enterprise RAG. Excellent citation generation and grounding. Multilingual at 10 languages. Cohere Coral SDK simplifies integration. Good for accuracy-critical search applications.
OpenAI's original cost-efficient chat model — now superseded
A 2023 model that costs five times what GPT-6 Luna does for a fraction of the quality and a 16K context. Avoid for new projects. Migrate to gpt-6-luna or a modern open-weight alternative.
