This comparison first ran in January 2025 with Qwen 2.5, DeepSeek R1, ChatGPT, and Claude 3.5 — all four are several generations behind now. Here’s what actually leads each lineup as of August 2026, and what they cost.
Qwen vs DeepSeek vs GPT-5.6 vs Claude: Quick Comparison (2026)
| Qwen 3.8-Max | DeepSeek V4-Pro | GPT-5.6 Sol | Claude Sonnet 5 | |
|---|---|---|---|---|
| Maker | Alibaba | DeepSeek | OpenAI | Anthropic |
| Released | Aug 3, 2026 | Aug 13, 2026 (V4 family: Apr 2026) | Jul 9, 2026 | 2026 (current gen) |
| Context window | 1M tokens | 1M tokens | Not fully disclosed | 1M tokens |
| Input price | $2.00 / MTok | $0.66 / MTok (off-peak, cache miss) | $5.00 / MTok | $2.00 / MTok |
| Output price | $6.00 / MTok | $1.98 / MTok (off-peak) | $30.00 / MTok | $10.00 / MTok |
| Notable | 2.4T params, open-weight variant followed a week later | Peak/off-peak pricing (2x, weekday mornings UTC) | 3-tier family: Sol/Terra/Luna | Launch pricing is now permanent — no Sept 2026 increase |
Sources: Qwen 3.8-Max specs, DeepSeek API pricing, GPT-5.6 pricing, Claude API pricing.
What Changed Since January 2025
Every model in the original comparison has been replaced at least twice:
- Qwen 2.5 → Qwen 3.8-Max. Alibaba moved through the Qwen 3 series before previewing 3.8-Max at the World AI Conference in Shanghai (July 19, 2026) and releasing it August 3, 2026 — 2.4 trillion parameters, native multimodality, and a 1M-token context window.
- DeepSeek R1 → DeepSeek V4. R1 gave way to R1-0528, then the V3.1/V3.2 hybrid-reasoning line, then the V4 family in April 2026. The current lineup is deepseek-v4-flash and deepseek-v4-pro (V4-Pro GA’d August 13, 2026) — there’s no “V4.2.”
- ChatGPT → GPT-5.6. The GPT-4-era model referenced in the original article gave way to the GPT-5 series, which now ships as three tiers — Luna, Terra, and Sol — released July 9, 2026.
- Claude 3.5 → Claude Sonnet 5. Anthropic’s lineup moved through the 4.x generation to the current Claude 5 family: Sonnet 5, Opus 5, and the limited-availability Fable 5 and Mythos 5.
Qwen 3.8-Max
Qwen 3.8-Max is Alibaba’s current flagship: 2.4 trillion parameters, a 1M-token context window, and native multimodality as the headline addition over the Qwen 3 line. It’s priced at $2 per million input tokens and $6 per million output tokens, and it’s available through OpenAI-compatible, Anthropic-compatible, and DashScope-native endpoints — useful if you’re already integrated against another vendor’s SDK shape. True to Alibaba’s usual pattern, a smaller open-weight sibling (Qwen3.8-2.4T-A95B) followed about a week after the hosted flagship launched.
DeepSeek V4
DeepSeek retired the R1 branding after R1-0528 (May 2025) and the V3.1/V3.2 hybrid-reasoning models that followed — see our full DeepSeek R1 retrospective for that lineage in detail. The current family is V4-Flash and V4-Pro, both with a 1M-token context window and 384K max output, both supporting configurable thinking mode. The unusual part is pricing: DeepSeek runs peak and off-peak rates, with peak hours set at 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, priced at 2x the off-peak rate. V4-Pro off-peak runs $0.66/MTok input (cache miss) and $1.98/MTok output — still dramatically cheaper than GPT-5.6 Sol or Claude Opus 5, even during peak hours.
GPT-5.6: Sol, Terra, and Luna
OpenAI replaced the single-model ChatGPT lineup with a three-tier GPT-5.6 family, released July 9, 2026: Luna (cheapest), Terra (balanced), and Sol (flagship, what Pro mode runs on for the hardest tasks) — see our full GPT-5.6 breakdown for how the tiers differ. OpenAI cut prices on July 30, 2026 — Luna dropped 80% (to $0.20/$1.20 per MTok) and Terra dropped 20% (to $2.00/$12.00) — while Sol held steady at $5.00/$30.00. Sol is the most expensive model in this comparison on both input and output.
Claude Sonnet 5 and Opus 5
Claude 3.5 is long retired. The current lineup is Sonnet 5 ($2/$10 per MTok), Opus 5 ($5/$25), Haiku 4.5 ($1/$5), plus the limited-availability Fable 5 and Mythos 5 ($10/$50). One pricing note worth flagging: Sonnet 5 launched with introductory pricing that was scheduled to rise to $3/$15 on September 1, 2026 — Anthropic cancelled that increase, so $2/$10 is now the permanent standard rate, not a promo that’s about to expire.
Verdict: Which Should You Use in 2026?
For raw cost efficiency, DeepSeek V4-Pro off-peak is the cheapest frontier-class model here by a wide margin — schedule batch workloads outside its Mon–Fri peak window and the savings compound. For a balance of price and reasoning quality, Qwen 3.8-Max and Claude Sonnet 5 land in the same $2 input tier but with different strengths: Qwen for native multimodal work and OpenAI/Anthropic-compatible endpoints, Sonnet 5 for coding and agentic tool use. GPT-5.6 Terra is the practical middle tier of OpenAI’s lineup now that Luna and Terra both got cheaper; reach for Sol only when you need the absolute ceiling on reasoning and can absorb $30/MTok output. Claude Opus 5 is the equivalent ceiling on the Anthropic side at roughly a third of Sol’s output cost.
Frequently Asked Questions
Q: Is Qwen 2.5 still Alibaba’s best model?
A: No. Alibaba has released multiple Qwen 3 generations since, and the current flagship is Qwen 3.8-Max (August 2026).
Q: Is DeepSeek R1 still DeepSeek’s current model?
A: No. R1 was succeeded by R1-0528, then the V3.1/V3.2 line, then the V4 family (V4-Flash and V4-Pro) that’s current as of August 2026.
Q: What happened to “ChatGPT” as a model name?
A: ChatGPT is the product; the underlying model has moved through several GPT-5 generations and now ships as the three-tier GPT-5.6 family (Luna, Terra, Sol), released July 2026.
Q: Is Claude 3.5 still current?
A: No. Claude 3.5 is retired. The current lineup is the Claude 5 family — Sonnet 5, Opus 5, Fable 5, and Mythos 5 — plus Haiku 4.5.
Q: Which model is cheapest for high-volume use?
A: DeepSeek V4-Pro off-peak, at $0.66/$1.98 per MTok — though its peak-hour pricing (2x, weekday mornings UTC) narrows that gap if your traffic doesn’t shift around it.
Quick Summary:
- Qwen 2.5, DeepSeek R1, and Claude 3.5 — the models in this article’s original 2025 version — are all retired
- Current flagships: Qwen 3.8-Max ($2/$6 per MTok), DeepSeek V4-Pro ($0.66/$1.98 off-peak), GPT-5.6 Sol ($5/$30), Claude Opus 5 ($5/$25) / Sonnet 5 ($2/$10)
- DeepSeek V4 uses peak/off-peak pricing — 2x cost Mon–Fri 01:00–04:00 and 06:00–10:00 UTC
- GPT-5.6 ships as three tiers (Luna, Terra, Sol) instead of one model; Luna and Terra got cheaper on July 30, 2026
- Claude Sonnet 5’s $2/$10 launch pricing is now permanent — the planned September 2026 increase to $3/$15 was cancelled